ࡱ>  9;678f5@ Rbjbj22 s[XX{::::MMM8M\PڙnPa"aaadddY[[[[[[$HRdkd^ddkdk::aawwwdk: aaYwdkYwwx=͎aP >Mfp %40ڙ#PuP4͎::::P͎X dfNwg  iYddd6FMDvFM Practical Selection of SVM Parameters and Noise Estimation for SVM Regression Vladimir Cherkassky and Yunqian Ma* Department of Electrical and Computer Engineering, University of Minnesota, Minneapolis, MN 55455, USA Abstract We investigate practical selection of meta-parameters for SVM regression (that is, EMBED Equation.3-insensitive zone and regularization parameter C). The proposed methodology advocates analytic parameter selection directly from the training data, rather than resampling approaches commonly used in SVM applications. Good generalization performance of the proposed parameter selection is demonstrated empirically using several low-dimensional and high-dimensional regression problems. Further, we point out the importance of Vapniks EMBED Equation.3-insensitive loss for regression problems with finite samples. To this end, we compare generalization performance of SVM regression (with optimally chosen EMBED Equation.3) with regression using least-modulus loss (EMBED Equation.3=0). These comparisons indicate superior generalization performance of SVM regression, for finite sample settings. Keywords: Complexity Control; Parameter Selection; Support Vector Machine; VC theory 1. Introduction This study is motivated by a growing popularity of support vector machines (SVM) for regression problems [3,6-14]. Their practical successes can be attributed to solid theoretical foundations based on VC-theory [13,14], since SVM generalization performance does not depend on the dimensionality of the input space. However, many SVM regression application studies are performed by expert users having good understanding of SVM methodology. Since the quality of SVM models depends on a proper setting of SVM meta-parameters, the main issue for practitioners trying to apply SVM regression is how to set these parameter values (to ensure good generalization performance) for a given data set. Whereas existing sources on SVM regression [3,6-14]  *Corresponding author. Email addresses:  HYPERLINK "mailto:cherkass@ece.umn.edu" cherkass@ece.umn.edu (V. Cherkassky),  HYPERLINK "mailto:myq@ece.umn.edu" myq@ece.umn.edu (Y. Ma) give some recommendations on appropriate setting of SVM parameters, there is clearly no consensus and (plenty of) contradictory opinions. Hence, resampling remains the method of choice for many applications. Unfortunately, using resampling for (simultaneously) tuning several SVM regression parameters is very expensive in terms of computational costs and data requirements. This paper describes simple yet practical analytical approach to SVM regression parameter setting directly from the training data. Proposed approach (to parameter selection) is based on well-known theoretical understanding of SVM regression that provides the basic analytical form of dependencies for parameter selection. Further, we perform empirical tuning of such dependencies using several synthetic data sets. Practical validity of the proposed approach is demonstrated using several low-dimensional and high-dimensional regression problems. Recently, several researchers [10,13,14] noted similarity between Vapniks EMBED Equation.3-insensitive loss function and Hubers loss in robust statistics. In particular, Vapniks loss function coincides with a special form of Hubers loss aka least-modulus loss (with EMBED Equation.3=0). From the viewpoint of traditional robust statistics, there is well-known correspondence between the noise model and optimal loss function [10]. However, this connection between the noise model and the loss function is based on (asymptotic) maximum likelihood arguments [10]. It can be argued that for finite sample regression problems Vapniks EMBED Equation.3-insensitive loss (with properly chosen EMBED Equation.3-parameter) actually would yield better generalization than other loss function (known to be asymptotically optimal for a particular noise density). In order to test this assertion, we compare generalization performance of SVM regression (with optimally chosen EMBED Equation.3) with robust regression using least-modulus loss function (EMBED Equation.3=0) for several noise densities. This paper is organized as follows. Section 2 gives a brief introduction to SVM regression and reviews existing methods for SVM parameter setting. Section 3 describes the proposed approach to selecting SVM regression parameters. Section 4 presents empirical comparisons demonstrating the advantages of the proposed approach. Section 5 describes empirical comparisons for regression problems with non-Gaussian noise; these comparisons indicate that SVM regression (with optimally chosen EMBED Equation.3) provides better generalization performance than SVM with least-modulus loss. Section 6 describes noise variance estimation for SVM regression. Finally, summary and discussion are given in Section 7. 2. Support Vector Regression and SVM Parameter Selection In regression formulation, the goal is to estimate an unknown continuous-valued function based on a finite number set of noisy samplesEMBED Equation.3, where d-dimensional inputEMBED Equation.3and the outputEMBED Equation.3 . Assumed statistical model for data generation has the following form: EMBED Equation.3 (1) where EMBED Equation.3is unknown target function (regression), and EMBED Equation.3 is additive zero mean noise with noise variance EMBED Equation.3[3,4]. In SVM regression, the inputEMBED Equation.3is first mapped onto a m-dimensional feature space using some fixed (nonlinear) mapping, and then a linear model is constructed in this feature space [3,10,13,14]. Using mathematical notation, the linear model (in the feature space) EMBED Equation.3is given by EMBED Equation.3 (2) where EMBED Equation.3denotes a set of nonlinear transformations, and b is the bias term. Often the data are assumed to be zero mean (this can be achieved by preprocessing), so the bias term in (2) is dropped. The quality of estimation is measured by the loss functionEMBED Equation.3. SVM regression uses a new type of loss function called EMBED Equation.3-insensitive loss function proposed by Vapnik [13,14]: EMBED Equation.3 (3) The empirical risk is: EMBED Equation.3 (4) Note that EMBED Equation.3-insensitive loss coincides with least-modulus loss and with a special case of Hubers robust loss function [13,14] when EMBED Equation.3=0. Hence, we shall compare prediction performance of SVM (with proposed chosen EMBED Equation.3) with regression estimates obtained using least-modulus loss(EMBED Equation.3=0) for various noise densities. SVM regression performs linear regression in the high-dimension feature space using EMBED Equation.3-insensitive loss and, at the same time, tries to reduce model complexity by minimizing EMBED Equation.3. This can be described by introducing (non-negative) slack variablesEMBED Equation.3 EMBED Equation.3, to measure the deviation of training samples outside EMBED Equation.3-insensitive zone. Thus SVM regression is formulated as minimization of the following functional: min EMBED Equation.3 s.t. EMBED Equation.3 (5) This optimization problem can transformed into the dual problem [13,14], and its solution is given by EMBED Equation.3 s.t. EMBED Equation.3,EMBED Equation.3, (6) where  EMBED Equation.2 is the number of Support Vectors (SVs) and the kernel function  EMBED Equation.2  (7) It is well known that SVM generalization performance (estimation accuracy) depends on a good setting of meta-parameters parameters C,  EMBED Equation.2  and the kernel parameters. The problem of optimal parameter selection is further complicated by the fact that SVM model complexity (and hence its generalization performance) depends on all three parameters. Existing software implementations of SVM regression usually treat SVM meta-parameters as user-defined inputs. In this paper we focus on the choice of C and  EMBED Equation.2 , rather than on selecting the kernel function. Selecting a particular kernel type and kernel function parameters is usually based on application-domain knowledge and also should reflect distribution of input (x) values of the training data [1,12,13,14]. For example, in this paper we show examples of SVM regression using radial basis function(RBF) kernels where the RBF width parameter should reflect the distribution/range of x-values of the training data. Parameter C determines the trade off between the model complexity (flatness) and the degree to which deviations larger than  EMBED Equation.2  are tolerated in optimization formulation (5). For example, if C is too large (infinity), then the objective is to minimize the empirical risk (4) only, without regard to model complexity part in the optimization formulation (5). Parameter  EMBED Equation.2 controls the width of the  EMBED Equation.2 -insensitive zone, used to fit the training data [3,13,14]. The value of  EMBED Equation.2 can affect the number of support vectors used to construct the regression function. The bigger  EMBED Equation.2 , the fewer support vectors are selected. On the other hand, bigger  EMBED Equation.2 -values result in more flat estimates. Hence, both C and  EMBED Equation.2 -values affect model complexity (but in a different way). Existing practical approaches to the choice of C and  EMBED Equation.2 can be summarized as follows: Parameters C and  EMBED Equation.2 are selected by users based on a priori knowledge and/or user expertise [3,12,13,14]. Obviously, this approach is not appropriate for non-expert users. Based on observation that support vectors lie outside the  EMBED Equation.2 -tube and the SVM model complexity strongly depends on the number of support vectors, Schlkopf et al [11] suggest to control another parameter EMBED Equation.2 (i.e., the fraction of points outside the  EMBED Equation.2 -tube) instead of  EMBED Equation.2 . Under this approach, parameter  EMBED Equation.2 has to be user-defined. Similarly, Mattera and Haykin [7] propose to choose  EMBED Equation.2 - value so that the percentage of support vectors in the SVM regression model is around 50% of the number of samples. However, one can easily show examples when optimal generalization performance is achieved with the number of support vectors larger or smaller than 50%. Smola et al [9] and Kwok [6] proposed asymptotically optimal  EMBED Equation.2 - values proportional to noise variance, in agreement with general sources on SVM [3,13,14]. The main practical drawback of such proposals is that they do not reflect sample size. Intuitively, the value of  EMBED Equation.2  should be smaller for larger sample size than for a small sample size (with the same level of noise). Selecting parameter C equal to the range of output values [7]. This is a reasonable proposal, but it does not take into account possible effect of outliers in the training data. Using cross-validation for parameter choice [3,12]. This is very computation and data-intensive. Several recent references present statistical account of SVM regression [10,5] where the  EMBED Equation.2 - parameter is associated with the choice of the loss function (and hence could be optimally tuned to particular noise density) whereas the C parameter is interpreted as a traditional regularization parameter in formulation (5) that can be estimated for example by cross-validation [5]. As evident from the above, there is no shortage of (conflicting) opinions on optimal setting of SVM regression parameters. Under our approach (described next in Section 3) we propose: - Analytical selection of C parameter directly from the training data (without resorting to resampling); - Analytical selection of  EMBED Equation.2 - parameter based on (known or estimated) level of noise in the training data. Further ample empirical evidence presented in this paper suggests the importance of  EMBED Equation.2 -insensitive loss, in the sense that SVM regression (with proposed parameter selection) consistently achieves superior prediction performance vs other (robust) loss functions, for different noise densities. 3. Proposed Approach for Parameter Selection Selection of parameter C. Optimal choice of regularization parameter C can be derived from standard parameterization of SVM solution given by expression (6):  EMBED Equation.2   EMBED Equation.2   EMBED Equation.2  (8) Further we use kernel functions bounded in the input domain. To simplify presentation, assume RBF kernel function  EMBED Equation.2  (9) so that EMBED Equation.2 . Hence we obtain the following upper bound on SVM regression function:  EMBED Equation.2  (10) Expression (10) is conceptually important, as it relates regularization parameter C and the number of support vectors, for a given value of  EMBED Equation.2 . However, note that the relative number of support vectors depends on the  EMBED Equation.2 -value. In order to estimate the value of C independently of (unknown)  EMBED Equation.2 , one can robustly let  EMBED Equation.2  for all training samples, which leads to setting C equal to the range of response values of training data [7]. However, such a setting is quite sensitive to the possible presence of outliers, so we propose to use instead the following prescription for regularization parameter:  EMBED Equation.2  (11) where  EMBED Equation.2  is the mean of the training responses (outputs), and  EMBED Equation.2  is the standard deviation of the training response values. Prescription (11) can effectively handle outliers in the training data. In practice, the response values of training data are often scaled so that  EMBED Equation.2 =0; then the proposed C is  EMBED Equation.2 . Selection of  EMBED Equation.2 . It is well-known that the value of  EMBED Equation.2  should be proportional to the input noise level, that is  EMBED Equation.2 [3,6,9,13]. Here we assume that the standard deviation of noise  EMBED Equation.2  is known or can be estimated from data (practical approaches to noise estimation are discussed in Section 6). However, the choice of  EMBED Equation.2 should also depend on the number of training samples. From standard statistical theory, the variance of observations about the trend line (for linear regression) is:  EMBED Equation.2  (12) This suggests the following prescription for choosing  EMBED Equation.2 :  EMBED Equation.2  (13) Based on a number of empirical comparisons, we found that (13) works well when the number of samples is small, however for large values of n prescription (13) yields  EMBED Equation.2 -values that are too small. Hence we propose the following (empirical) dependency:  EMBED Equation.2  (14) Based on empirical tuning, the constant value  EMBED Equation.2  gives good performance for various data set sizes, noise levels and target functions for SVM regression. Thus expression (14) is used in all empirical comparisons presented in Sections 4 and 5. 4. Experimental Results for Gaussian Noise First we describe experimental procedure used for comparisons, and then present empirical results. Training data: simulated training data EMBED Equation.2  where x-values are sampled on uniformly-spaced grid in the input space, and y-values are generated according to EMBED Equation.2 . Different types of the target functions EMBED Equation.2 are used. The y-values of training data are corrupted by additive noise. We used Gaussian noise (results described in this section) and several non-Gaussian additive symmetric noise densities (discussed in Section 5). Since SVM approach is not sensitive to a particular noise distribution, we expect to show good generalization performance with different types of noise, as long as an optimal value of  EMBED Equation.2  (reflecting standard deviation of noise  EMBED Equation.2 ) has been used. Test data: the test inputs are sampled randomly according to uniform distribution in x-space. Kernel function: RBF kernel functions (9) are used in all experiments, and the kernel width parameter p is appropriately selected to reflect the input range of the training/test data. Namely, the RBF width parameter is set to p ~ (0.2-0.5)* range (x). For higher d-dimensional problems the RBF width parameter is set so that pd ~ (0.2-0.5) where all d input variables are pre-scaled to [0,1] range. Such values yield good SVM performance for various regression data sets. Performance metric: since the goal is optimal selection of SVM parameters in the sense of generalization, the main performance metric is prediction risk  EMBED Equation.2  (15) defined as MSE between SVM estimates and true values of the target function for test inputs. The first set of results show how SVM generalization performance depends on a proper choice of SVM parameters for univariate sinc target function:  EMBED Equation.2  (16) The following values of  EMBED Equation.2 were used  EMBED Equation.2 to generate five data sets using small sample size (n=30) with additive Gaussian noise (with different noise levels  EMBED Equation.2  as shown in Table 1) . For these data sets, we used RBF kernels with width parameter p=4. Table 1 shows: (a) Parameter values C and  EMBED Equation.2  (using expressions proposed in Section 3) for different training sets. (b) Prediction risk and percentage of support vectors (%SV) obtained by SVM regression with proposed parameter values. (c) Prediction risk and percentage of support vectors (%SV) obtained using least-modulus loss function ( EMBED Equation.2 =0). We can see that the proposed method for choosing  EMBED Equation.2 is better than least-modulus loss function, as it yields lower prediction risk and better (more sparse) representation. Table 1 Results for univariate sinc function (small size): Data Set 1- Data Set5 Data Set  EMBED Equation.3  Noise C-selection  EMBED Equation.2 -selection Prediction %SV Level( EMBED Equation.3 ) Risk 1 1 0.2 1.58  EMBED Equation.2 =0 0.0129 100%  EMBED Equation.2 =0.2 (prop.) 0.0065 43.3% 2 10 2 15  EMBED Equation.2 =0 1.3043 100%  EMBED Equation.2 =2.0 (prop.) 0.7053 36.7% 3 0.1 0.02 0.16  EMBED Equation.2 =0 1.03e-04 100%  EMBED Equation.2 =0.02 (prop.) 8.05e-05 40.0% 4 -10 0.2 14.9  EMBED Equation.2 =0 0.0317 100%  EMBED Equation.2 =0.2 (prop.) 0.0265 50.0% 5 -0.1 0.02 0.17  EMBED Equation.2 =0 1.44e-04 100%  EMBED Equation.2 =0.02 (prop.) 1.01e-04 46.7%  Fig. 1. For Data Set 1, SVM estimate using proposed parameter selection vs using least-modulus loss. Visual comparisons (for univariate sinc Data Set 1) between SVM estimates using proposed parameter selection and using least-modulus loss are shown in Fig.1, where the solid line is the target function, the + denote training data, the dotted line is an estimate using least-modulus loss and the dashed line is the SVM estimate function using our method. The accuracy of expression (14) for selecting optimal  EMBED Equation.2 as a function of n (the number of training samples is demonstrated in Fig. 2. Results in Fig.2 show that proposed  EMBED Equation.2 -values vs optimal  EMBED Equation.2 -values (obtained by exhaustive search in terms of prediction risk) for Data Set 1 (see Table 1) for different number of training samples.  Fig. 2. Proposed  EMBED Equation.2 -values vs optimal  EMBED Equation.2 -values (obtained by exhaustive search in terms of prediction risk) for Data set 1 for different number of training data (n=30, 50, , 150). Dependence of prediction risk as a function of chosen C and  EMBED Equation.2 -values for Data Set 1 (i.e., sinc target function, 30 training samples) in shown in Fig. 3a. Fig. 3b shows the percentage of support vectors (%SV) selected by SVM regression, which is an important factor affecting generalization performance. Visual inspection of results in Fig.3a indicates that proposed choice of  EMBED Equation.2 , C gives good/ near optimal performance in terms of prediction risk. Also, one can clearly see that C-values above certain threshold have only minor effect on the prediction risk. Our method guarantees that the proposed chosen C-values result in SVM solutions in flat regions of prediction risk. Using three dimensional Fig.3b, we can see that small  EMBED Equation.2 -values correspond to higher percentage of support vectors, whereas parameter C has negligible effect on the percentage of SV selected by SVM method. Fig. 4 shows prediction risk as a function of chosen C and  EMBED Equation.2 -values for sinc target function for Data Set 2 and Data Set 3. We can see that the proposed choice of C yields optimal and robust C-value corresponding to SVM solutions in flat regions of prediction risk.  (a)  (b) Fig. 3. Results for small sample size, sinc target function: Set 1 (a) Prediction risk (b) The number of SV as a fraction of training data  (a)  (b) Fig. 4. Results for small sample size, sinc target function: (a) Prediction Risk for Data Set 2 (b) Prediction Risk for Data Set 3 In order to investigate the effect of the sample size (on selection of  EMBED Equation.2 -value), we generate 200 training samples using univariate sinc target function (as in Data Set 1) with Gaussian noise ( EMBED Equation.2 =0.2). Fig.5 shows the dependence of prediction risk on SVM parameters for this data set (large sample size). According to proposed expression (14) and (11), proposed  EMBED Equation.2 is 0.1, proposed C is 1.58, which is consistent with results in Fig.5. Also, the prediction risk is 0.0019, which compares favorably with SVM using least-modulus loss ( EMBED Equation.2 =0) where the prediction risk is 0.0038. Similarly, the proposed method compares favorably with selection EMBED Equation.2 proposed by Kwok [6]. For this data set, Kwoks method yields  EMBED Equation.2 and the prediction risk is 0.0033. The reason that our approach to  EMBED Equation.2 -selection gives better results is that previous methods for selecting  EMBED Equation.2 -value [6,9] do not depend on sample size.  Fig. 5. Result for large sample size sinc function (Data Set 1): Prediction risk Next we show results of SVM parameter selection for high-dimensional problems. The first data set is generated using two-dimensional sinc target function [13,14]  EMBED Equation.2  (17) defined on a uniform square lattice [-5,5]*[-5,5] , with response values corrupted with Gaussian noise ( EMBED Equation.2 =0.1 and  EMBED Equation.2  =0.4 respectively). The number of training samples is 169, and the number of test samples is 676. The RBF kernel width parameter p=2 is used. Fig. 6a shows the target function and Fig. 6b shows the SVM estimate obtained using proposed parameter selection for  EMBED Equation.2  =0.1. The proposed C =1.16 and  EMBED Equation.2 =0.05 (for  EMBED Equation.2  =0.1) and  EMBED Equation.2 =0.21 (for  EMBED Equation.2  =0.4). Table 2 compares the proposed parameter selection with estimates obtained using least-modulus loss, in terms of prediction risk and the percentage of SV chosen by each method.  (a)  (b) Fig. 6. (a) 2D sinc target function (b) SVM regression estimate using proposed method for for  EMBED Equation.2  =0.1 Table 2 Comparison of the proposed method for  EMBED Equation.2 with least-modulus loss ( EMBED Equation.2 =0) for two-dimensional sinc target function data sets. Noise  EMBED Equation.2 -selection Prediction %SV Level Risk  EMBED Equation.2 =0.1  EMBED Equation.2 =0 0.0080 100%  EMBED Equation.2  (Proposed) 0.0020 62.7%  EMBED Equation.2 =0.4  EMBED Equation.2 =0 0.0369 100%  EMBED Equation.2  (Proposed) 0.0229 60.9% Next we show results of SVM parameter selection for higher dimensional additive target function  EMBED Equation.2  (18) where  EMBED Equation.2 -values are distributed in hypercube  EMBED Equation.2 . Output (response) values of training samples are corrupted by additive Gaussian noise is (with  EMBED Equation.2 =0.1 and  EMBED Equation.2 =0.2). Training data size is n=243 samples (i.e., 3 points per each input dimension). The test size is 1024. The RBF kernel width parameter p=0.8 is used for this data set.. The optimal value of C is 34 and the optimal  EMBED Equation.2 =0.045 for  EMBED Equation.2  =0.1 and  EMBED Equation.2 =0.09 for  EMBED Equation.2  =0.2. Comparison results between the proposed methods for parameter selection with the method using least-modulus loss function are shown in Table 3. Clearly, the proposed approach gives better performance in terms of prediction risk and robustness. Table 3 Comparison of proposed method for  EMBED Equation.2 parameter selection with least-modulus loss ( EMBED Equation.2 =0) for high-dimensional additive target function Noise  EMBED Equation.2 -selection Prediction %SV Level Risk  EMBED Equation.2 =0.1  EMBED Equation.2 =0 0.0443 100%  EMBED Equation.2  (Proposed) 0.0387 86.7%  EMBED Equation.2 =0.2  EMBED Equation.2 =0 0.1071 100%  EMBED Equation.2  (Proposed) 0.0918 90.5% 5. Experimental Results for Non-Gaussian Noise This section describes empirical results for regression problems with non-Gaussian additive symmetric noise in the statistical model (1). The main motivation is to demonstrate practical advantages of Vapniks  EMBED Equation.2 -insensitive loss vs other loss functions. Whereas practical advantages of SVM regression are well-known, there is a popular opinion [10] that one should use a particular loss function for a given noise density. Hence, we perform empirical comparisons between SVM regression (with proposed parameter selection) vs SVM regression using least-modulus loss, for several finite-sample regression problems. First, consider Students t-distribution for noise. Univariate sinc target function is used for comparisons:  EMBED Equation.3 . Training data consists of n=30 samples. RBF kernels with width parameter p=4 are used for this data set. Several experiments have been performed using various degrees of freedom (5, 10, 20, 30, 40) for generating t-distribution. Empirical results indicate superior performance of the proposed method for parameter selection. Table 4 shows comparison results with least-modulus loss for Students noise with 5 degrees of freedom (when  EMBED Equation.2  of noise is 1.3). According to proposed expressions (14) and (11), proposed  EMBED Equation.2 is 1.3 and C is 16. Table 4 Comparison of proposed method for  EMBED Equation.2 with least-modulus loss ( EMBED Equation.2 =0) for t-distribution noise  EMBED Equation.2 -selection Prediction %SV Risk  EMBED Equation.2 =0 0.9583 100%  EMBED Equation.2  (Proposed) 0.6950 40% Next, we show comparison results for Laplacian noise density:  EMBED Equation.2  (19) Smola et al [10] suggest that for this noise density model, the least-modulus loss should be used. Whereas this suggestion might work in an asymptotical setting, it does not guarantee superior performance with finite samples. We compare the proposed approach for choosing  EMBED Equation.2 with the least-modulus loss method in noise density model (19). This experiment uses the same sinc target function as in Table 4 (with sample size n=30). The  EMBED Equation.2  of noise for Laplacian noise model (19) is 1.41 (precisely EMBED Equation.2 ). Using our proposed approach,  EMBED Equation.2 =1.41 and C is 16. Table 5 shows comparison results. Visual comparison of results in Table 5 is also shown in Fig. 7, where the solid line is the target function, the + denote training data, the dotted line is an estimate found using least-modulus loss and the dashed line is an estimate found using SVM method with proposed parameter selection. Table 5 Comparison of the proposed method for  EMBED Equation.2 with least-modulus loss ( EMBED Equation.2 =0) for Laplacian noise.  EMBED Equation.2 -selection Prediction %SV Risk  EMBED Equation.2 =0 0.8217 100%  EMBED Equation.2  (Proposed) 0.5913 46.7%  Fig. 7. SVM estimate using proposed parameter selection vs using least-modulus loss Finally, consider uniform distribution for the additive noise. Univariate sinc target function is used for comparisons: EMBED Equation.3 . Several experiments have been performed using different noise level  EMBED Equation.2 . Training sample size n=30 is used in the experiments. According to proposed expressions (14) and (11), C is 1.6,  EMBED Equation.2 is 0.1(for  EMBED Equation.2 =0.1),  EMBED Equation.2 is 0.2(for  EMBED Equation.2 =0.2),  EMBED Equation.2 is 0.3(for  EMBED Equation.2 =0.3). Table 6 shows comparison results. Table 6 Comparison of proposed method for  EMBED Equation.2 with least-modulus loss ( EMBED Equation.2 =0) for uniformly distributed noise Noise  EMBED Equation.2 -selection Prediction %SV Level Risk  EMBED Equation.2 =0.1  EMBED Equation.2 =0 0.0080 100%  EMBED Equation.2 (Proposed) 0.0036 60%  EMBED Equation.2 =0.2  EMBED Equation.2 =0 0.0169 100%  EMBED Equation.2 (Proposed) 0.0107 43.3%  EMBED Equation.2 =0.3  EMBED Equation.2 =0 0.0281 100%  EMBED Equation.2 (Proposed) 0.0197 50% 6. Noise Variance Estimation The proposed method for selecting  EMBED Equation.2 relies on the knowledge of the standard deviation of noise  EMBED Equation.2 . The problem, of course, is that the noise variance is not known a priori, and it needs to be estimated from training data EMBED Equation.2 . In practice, the noise variance can be readily estimated from the squared sum of residuals (fitting error) of the training data. Namely, the well-known approach of estimating noise variance (for linear models) is by fitting the data using low bias (high-complexity) model (say high-order polynomial) and applying the following formula to estimate noise [3,4]  EMBED Equation.2  (20) where  EMBED Equation.2  is the degrees of freedom (DOF) of the high-complexity estimator and n is the number of training samples. Note that for linear estimators (i.e., polynomial regression) DOF is simply the number of free parameters (polynomial degree); whereas the notion of DOF is not well defined for other types of estimators [3]. We used expression (20) for estimating noise variance using higher-order algebraic polynomials (for univariate regression problems) and k-nearest-neighbors regression. Both approaches yield very accurate estimates of the noise variance; however, we only show the results of noise estimation using k-nearest-neighbors regression. In k-nearest-neighbors method, the function is estimated by taking a local average of the training data. Locality is defined in terms of the k data points nearest the estimation point. The model complexity (DOF) of the k-nearest neighbors method can be estimated as:  EMBED Equation.2  (21) Even though the accuracy of estimating DOF for k-nearest-neighbors regression via (21) may be questionable, it provides rather accurate noise estimates when used in conjunction with (20). Combining expressions (20) and (21), we obtain the following prescription for noise variance estimation via k-nearest neighbors method:  EMBED Equation.2   EMBED Equation.2   EMBED Equation.2  (22) Typically, small values of k (in the 2-6 range) corresponding to low-bias/high variance estimators should be used in formula (22). In order to illustrate the effect of different k-values on the accuracy of noise variance estimation, we use three-dimension figure showing estimated noise as a function of k and n (number of training samples). Fig. 8 shows noise estimation results for univariate sinc target function corrupted by Gaussian noise with  EMBED Equation.2 =0.6 (noise variance is 0.36). It is evident from Fig. 8 that k-nearest neighbor method provides robust and accurate noise estimates with k-values chosen in a (2-6) range.  Fig.8. Using k-nearest neighbor method for estimating noise variance for univariate sinc function with different k and n values when the true noise variance =0.36 Since accurate estimation of noise variance does not seem to be affected much by specific k-value, we use k nearest neighbor method (with k=3). With k=3 expression (22) becomes  EMBED Equation.2 . (23) We performed noise estimation experiments using k-nearest neighbor method (with k=3) with different target functions, different sample size and different noise levels. In all cases, we obtained accurate noise estimates. Here, we only show noise estimation results for the univariate sinc target function for different true noise levels  EMBED Equation.2 =0.1, 0.2, 0.3, 0.4, 0.5, 0.6, 0.7, 0.8 (true noise variance is 0.01, 0.04, 0.09, 0.16, 0.25, 0.36, 0.49, 0.64 accordingly). Fig. 9 shows the scatter plot of noise level estimates obtained via (23) for 10 independently generated data sets (for each true noise level). Results in Fig.9 correspond to the least favorable experimental set-up for noise estimation (that is, small number of samples n=30 and large noise levels).  Fig. 9. Scatter plot of noise estimates obtained using k-nearest neighbors method (k=3) for univariate sinc function for different noise level (n=30) Empirical results presented in this section show how to estimate (accurately) the noise level from available training data. Hence, this underscores practical applicability of proposed expression (14) for  EMBED Equation.2 -selection. In fact, empirical results (not shown here due to space constraints) indicate that SVM estimates obtained using estimated noise level for  EMBED Equation.2 -selection yield similar prediction accuracy (within 5%) to SVM estimates obtained using known noise level, for data sets in Section 4 and 5. 7. Summary and Discussion This paper describes practical recommendations for setting meta-parameters for SVM regression. Namely the values of EMBED Equation.3and C parameters are obtained directly from the training data and (estimated) noise level. Extensive empirical comparisons suggest that the proposed parameter selection yields good generalization performance of SVM estimates under different noise levels, types of noise, target functions and sample sizes. Hence proposed approach for SVM parameter selection can be immediately used by practitioners interested in applying SVM to various application domains. Our empirical results suggest that with proposed choice of EMBED Equation.3, the value of regularization parameter C has only negligible effect on the generalization performance (as long as C is larger than a certain threshold analytically determined from the training data). The proposed value of C-parameter is derived for RBF kernels; however the same approach can be applied to other kernels bounded in the input domain. For example, we successfully applied proposed parameter selection for SVM regression with polynomial kernel defined in  EMBED Equation.2 (or EMBED Equation.2 ) input domain. Future related research may be concerned with investigating optimal selection of parameters C and EMBED Equation.3 for different kernel types, as well as optimal selection of kernel parameters (for these types of kernels). In this paper (using RBF kernels), we used fairly straightforward procedure for a good setting of RBF width parameter independent of C and EMBED Equation.3selection, thereby conceptually separating kernel parameter selection from SVM meta-parameter selection. However, it is not clear whether such a separation is possible with other kernel types. The second contribution of this paper is demonstrating the importance of EMBED Equation.3- insensitive loss function on the generalization performance. Several recent sources [10,5] assert that an optimal choice of the loss function (i.e. least-modulus loss, Hubers loss, quadratic loss etc.) should match a particular type of noise density (assumed to be known). However, these assertions are based on proofs asymptotic in nature. So we performed a number of empirical comparisons between SVM regression (with optimally chosen parameter values) and least-modulus regression (with EMBED Equation.3=0). All empirical comparisons show that SVM regression with EMBED Equation.3- insensitive loss provide better prediction performance that regression with least-modulus loss, even in the case of Laplacian noise (for which least-modulus regression is known to be statistically optimal). Likewise, recent study [2] shows that SVM loss (with proposed EMBED Equation.3) outperforms other commonly used loss functions (squared loss, least-modulus loss) for linear regression with finite samples. Intuitively, superior performance of EMBED Equation.3-insensitive loss for finite-sample problems can be explained by noting that noisy data samples very close to the true target function should not contribute to the empirical risk. This idea is formally reflected in Vapniks loss function, whereas Hubers loss function assigns squared loss to samples with accurate (close to the truth) response values. Conceptually, our findings suggest that for finite-sample regression problems we only need the knowledge of noise level (for optimal setting of EMBED Equation.3), instead of the knowledge of noise density. In other words, optimal generalization performance of regression estimates depends mainly on the noise variance rather than noise distribution. The noise variance itself can be estimated directly from the training data, i.e. by fitting very flexible (high-variance) estimator to the data. Alternatively, one can first apply least-modulus regression to the data, in order to estimate noise level. Further research in this direction may be needed, to gain better understanding of the relationship between optimal loss function, noise distribution and the number of training samples. In particular, an interesting research issue is to find the minimum number of samples beyond which a theoretically optimal loss function (for a given noise density) indeed provides superior generalization performance. Acknowledgements The authors thank Dr. V. Vapnik for many useful discussions. We also acknowledge several useful suggestions from anonymous reviewers. This work was supported, in part, by NSF grant ECS-0099906. References [1] O. Chapelle and V. Vapnik,Model Selection for Support Vector Machines. In Advances in Neural Information Processing Systems, Vol 12, (1999) [2] V. Cherkassky and Y. Ma, Selecting of the Loss Function for Robust Linear Regression, Neural computation, under review (2002) [3] V. Cherkassky and F. Mulier, Learning from Data: Concepts, Theory, and Methods. (John Wiley & Sons, 1998) [4] V. Cherkassky, X. Shao, F.Mulier and V. Vapnik, Model Complexity Control for Regression Using VC Generalization Bounds. IEEE Transaction on Neural Networks, Vol 10, No 5 (1999) 1075-1089 [5] T. Hastie, R. Tibshirani and J. Friedman, The Elements of Statistical Learning. Data Mining, Inference and Prediction, (Springer, 2001) [6] J.T. Kwok, Linear Dependency between  EMBED Equation.2  and the Input Noise in  EMBED Equation.2 Support Vector Regression, in: G. Dorffner, H. Bishof, and K. Hornik (Eds): ICANN 2001, LNCS 2130 (2001) 405-410 [7] D. Mattera and S. Haykin, Support Vector Machines for Dynamic Reconstruction of a Chaotic System, in: B. Schlkopf, J. Burges, A. Smola, ed., Advances in Kernel Methods: Support Vector Machine, MIT Press, (1999) [8] K. Muller, A. Smola, G. Ratsch, B. Scholkopf, J. Kohlmorgen, V. Vapnik, Using Support Vector Machines for Time Series Prediction, in: B. Scholkopf, J. Burges, A. Smola, ed., Advances in Kernel Methods: Support Vector Machine, MIT Press, (1999) [9] A. Smola, N. Murata, B.Schlkopf and K. Muller, Asymptotically optimal choice of  EMBED Equation.2 loss for support vector machines, Proc. ICANN, (1998) [10] A.Smola and B.Schlkopf. A Tutorial on Support Vector Regression. NeuroCOLT Technical Report NC-TR-98-030, Royal Holloway College, University of London, UK, 1998 [11] B.Schlkopf, P.Bartlett, A.Smola, and R.Williamson. Support Vector regression with automatic accuracy control, in L.Niklasson, M.Bodn, and T.Ziemke, ed., Proceedings of ICANN'98, Perspectives in Neural Computing, (Springer, Berlin, 1998) 111-116 [12] B. Scholkopf, J. Burges, A. Smola, Advances in Kernel Methods: Support Vector Machine. (MIT Press, 1999) [13] V. Vapnik. The Nature of Statistical Learning Theory (2nd ed.). (Springer, 1999) [14] V. Vapnik. Statistical Learning Theory, (Wiley, New York, 1998) Vladimir Cherkassky is with Electrical and Computer Engineering at the University of Minnesota. He received Ph.D. in Electrical Engineering from University of Texas at Austin in 1985. His current research is on methods for predictive learning from data, and he has co-authored a monograph Learning From Data published by Wiley in 1998. He has served on editorial boards of IEEE Transactions on Neural Networks, the Neural Networks Journal, and the Neural Processing Letters. He served on the program committee of major international conferences on Artificial Neural Networks, including International Joint Conference on Neural Networks (IJCNN), and World Congress on Neural Networks (WCNN). He was Director of NATO Advanced Study Institute (ASI) From Statistics to Neural Networks: Theory and Pattern Recognition Applications held in France, in 1993. He presented numerous tutorials and invited lectures on neural network and statistical methods for learning from data. Yunqian Ma is PhD candidate in Department of Electrical Engineering at the University of Minnesota. He received M.S. in Pattern Recognition and Intelligent System at Tsinghua University, P.R.China in 2000. His current research interests include support vector machines, neural network, model selection, multiple model estimation, and motion analysis. PAGE  PAGE 1 QRtuv> ? O P Q R    ܖ{iX{L{hth6CJaJ!jhthCJEHUaJ#j @ hthCJUVaJjhthCJUaJhth'(5CJaJhthCJaJ hGhgx*hhgx*6CJaJh /h6CJaJhCJaJhOCJaJhMhCJaJ hGhhMh5CJ aJ hgx*hgx*h 5CJ aJ >QRv} ~ $ H!]a$gd'( $h`ha$gd$ !]a$gdG`gdGgdG &dPgdG $L]La$gdGgdgx* $da$gdG       } ~ °|ui^SKCSKC^hKmCJaJhyCJaJhthKmCJaJhthCJaJhth6CJaJ hGhQw!jhthCJEHUaJ#j @ hthCJUVaJ!jhthCJEHUaJ#j @ hthCJUVaJhthCJaJjhthCJUaJ!jhthCJEHUaJ#j @ hthCJUVaJ Q R S T V W '()¾····¾··³}k}#jhOhOCJUaJjhOhOCJUaJhOhO6CJaJhOhOCJaJjhOUmHnHtHuhO hGhHh9h hGhhGh5hGhEH hGhgx*hhGh6hthCJaJhyCJaJ"  -NO $h`ha$gdD $`a$gdG$a$gdO $h`ha$gd?~dgdGdgdGgdG$&dPa$gdG)=>PQuvw LNPQSTxy?@PQRS޹ӵtm_Ttj5 hGh9hUj @ hGh9hUV hGh9hjhGh9hUj> hGh9hUj @ hGhUVjhGhU hGhHh hGhhOhOCJaJ#j{hOhOCJUaJhOhOCJaJjhOhOCJUaJ$hOhO0J>*B*CJaJph SfhUVfghipqrøxmxb^xhDjhGh%KUjhGh%KUj @ hGh%KUV hGh%KjhGh%KUj#hGh9hUj @ hGh9hUV hGh9hjhGh9hUh9hj, hGhUj @ hGhUV hGhjhGhU hGhH hGh#O !"#+,>?OPQR`aqrst⼯╈zmiW#j@ hGhdUVnHtHhDjhGhEHUj @ hGhUVjuhGhFUEHU#j@ hGhFUUVnHtHhGh6jhGhFUEHU#j@ hGhFUUVnHtHjhGhUhGhEH hGhjhGh%KUjhGh%KUEFVWXYʸzmibiWP hGhdjhGhdU hGhHhQj#hGhEHUj @ hGhUVj!hGhEHUj @ hGhUVhGh6jhGhdEHU#j8@ hGhdUVnHtH hGhFUjhGhFUUhD hGhjhGhUjhGhdEHU? h ;!r!!!l#[%v%%K&&$^`a$gd_$ !4$]a$gd_$^`a$gd_ $`a$gdG $4$a$gd3 $h`ha$gdrz$a$gdG $4$a$gdX`1 $h`ha$gdjWgdG $4$a$gdD./ !1234dλλλΰΞΰ|oΰ]Pj,hGh ,EHU#j@ hGh ,UVnHtHj)hGh-EEHU(j@ h-ECJUVaJnHo(tHj'hGh ,EHU#j~@ hGh ,UVnHtHjhGhU hGhHhX`1hGh6 hGhjhGhdUj%hGh(EHU#j^@ hGh(UVnHtH_ !!!!3!4!6!7!;!$?$@$$$$$$$$$$$$$$$$$$$[%a%b%r%s%t%u%⹬➑xtbUjLKhGhEHU#j8@ hGhUVnHtHh_jcIhGhUj @ hGhUVjAGhGhK6EHUj @ hGhUVjEhGhK6EHUj @ hGhUVjBhGh`]EHUj @ hGhUV hGhjhGhUjAhGhU!u%w%x%y%z%}%~%%%%%%%%%%%%&&&(&)&K&L&\&]&^&_&c&d&e&f&g&h&i&y&z&{&|&}&~&&&p#j$@ hGh yUVnHtHjQThGh yEHU#j$@ hGh yUVnHtHjmQhGhEHU#jQ@ hGhUVnHtH hGhHh_j"NhGhEHU#j=@ hGhUVnHtHjhGhUh_ hGh)&&&&&&&&&&& ' ''' '!'D'e''''' ((((((v)x)|)})))))d*ʽ૞yrgj}_hGhU hGhj]hGhUj4!? hGhUVhGh6hUjZhGhEHU#jc@ hGhUVnHtHjXhGhEHU#jo@ hGhUVnHtHh_ hGhjhGhUjVhGh yEHU%& 't'_+,.J/34^55M7N78o88':(:U:V:dgdG$H\^H`\a$gdG$a$gdG $ & Fa$gdG`gdG $h`ha$gdFg $`a$gdG $4$a$gdUgdGd*e********************?+@+i+j+++++++1,2,,,,,,,--%-&-'-(-Z-\-]-^-`-a-q-r---𥁄𯙽𯎽hEj8ehGhUjOchGhUjfahGhUj4!? hGhUVjhGhUhGh6 hGhh n hGhkp hGhH hGhW hGhhGh56---------@.A.S.T.U.V.........////)/*/+/,/U/V/[/\/n/o/p/q/////////C0D0V0W0ԾԫԠԕ hGhHhEjphGhUjnhGhUjlhGhUhGh6jjhGhUj ihGhUj4!? hGhUV hGhjhGhUj!ghGhU2W0X0Y000000000(1)1;1<1=1>1P1Q1c1d1e1f11111111111111133(3)3ɼ⮣⮘yngg hGhHj{hGhUj zhGhEHU#j< @ hGhUVnHtHj7xhGhUjNvhGhUj4!? hGhUVjthGhEHU#j2 @ hGhUVnHtH hGh7 hGhjhGhUjrhGhU()3K3L3^3_3`3a3333333/404B4C4D4E4444455556 6 6 666+6,6-6.6I7J7M7N7888 8!8o8q8{8888888B9C9U9V9jhGhU hGhgh hGh{8jhGhUhGh6jhGhU hGhHh j}hGhUj4!? hGhUVjhGhU hGh:V9W9X9':(:U:V:n:::; ; ; ; ;; ;!;3;4;5;6;7;8;J;K;L;M;[;_;x;;;;;<<⳦␃qd]] hGh Aj>hGhEHU#j@ hGhUVnHtHjVhGh"xEHU#j7@ hGh"xUVnHtHhjdhGh"xEHU#j@ hGh"xUVnHtHhGh6hGhEHhGh A5EH\ hGhjhGhUj{hGhU$V:: ;7;;<z<<a=@d@A+DDDE $4$a$gd"gdG $4$a$gd9# $4$a$gd $`a$gdG $4$a$gdA$a$gdG $4$`a$gd $4$]a$gd$a$gd$^`a$gd $h`ha$gd%e<<(<)<*<+<G<P<Q<<<<<<<<<<<<<#=B=========>>>>홌yrdYjrhGhUj4!? hGhUV hGh; hGhhhGh6hAjhGhxcEHU#j@ hGhxcUVnHtHjhGhhEHU#j@ hGhhUVnHtHhjhGhhEHU#j@ hGhhUVnHtH hGhjhGhU">N>O>a>b>c>d>>>>>>>>>>>>>>>>>> ?"?Z?[?@@@@@@/@zeXjhGhAmEHU(jj@ hAmCJUVaJnHo(tH hGhHjAhGhxcEHU#jA@ hGhxcUVnHtH hGhHmjDhGh'YEHU#jŧ$@ hGh'YUVnHtH hGhxchGh6j[hGhUj4!? hGhUVjhGhU hGh"/@M@i@j@k@}@~@@@@@@@@@AAAAAAAAAAAAAAA~qiW#jx"@ hGhUVnHtHhGh6j4hGhiwEHU(jk@ hiwCJUVaJnHo(tH hGhiwjhGhiwUj=hGhEHU#j!@ hGhUVnHtHj=hGhs-EHU(jk@ hs-CJUVaJnHo(tHjhGhUhiw hGhhAAAAAABBBB,B-B?B@BABBB|B}BBBBBBBBBBBBBBBBBػࠕvrkkk]j(uY? hGhUV hGhHhj hGhEHU#j"@ hGhUVnHtHj hGhUj4!? hGhUVj7hGh6Uj4!? hGh6UVjhGh6UhGh6 hGhjhGhUj4hGhEHU!BBBnCoCCCCC4D5DGDHDIDJDqDDDDDDDDDEEEEEEF?FⷪԔuqmbjhhGhUhKyh"jhGh-2EHU#j"@ hGhUVnHtHjhGhsU hGh 5h9#jhGh-2EHU#j"@ hGhUVnHtHjhGhUj4!? hGhUV hGhjhGhUjhGhU&EF GHH?H@HHKKMlNNOOO $`a$gdGh`hgd=$a$gdG $4$`a$gder $`a$gdG $h`ha$gd5 $h`ha$gd $h`ha$gdM}gdGdgdG $4$a$gd-2 $h`ha$gd-2?FFFFFFFFF9G:G;GMGNGOGPGQGHH?HHHHHHHHHHH,I-IOIPIbI|og_hh6haChaC5jhGhbzEHU#j`@ hGhbzUVnHtHhGh6hGhEHhGh5EHj;hGh-2EHU(jo@ h-2CJUVaJnHo(tHh-2jQhGh-2EHU#j#@ hGhUVnHtHjhGhU hGh"bIcIdIeIIIIIIIIII6K7KIKJKKKLKuKvKKKKKKKKKK L L$L`LaLLLμΫΕ|qiai]iih5hGh5hGh6jhGhUj(uY? hGhUVjhGhUj4!? hGhUVhh6hjhGhv2EHU#j@ hGhv2UVnHtH hGhjhGhUjehGhv2EHU#j@ hGhv2UVnHtH$LMM?M@MAMNMXMYMqMrMMMlNmNNNNNNNOOOOOOOOOOPPPPPPP0Pƹwsha hGhiwjhGhiwUhN!j.hGhEHU#j$@ hGhUVnHtHjhGhcgrEHU#jE@ hGhUVnHtHherjmhGhiwEHU#j@ hGhiwUVnHtHjhGhUhThGh 6H*hGh6 hGh&0P1P2P3P4P?PCPPPPPPPPPPPPPPPPQQ,Q-Q=Q>QCQDQVQȽ~vnf_fTjhGhU hGhehGh6hcgrCJaJh-ECJaJjQh^`hcgrCJUaJ#j(uY? h^`hcgrCJUVaJh^`hcgrCJaJjh^`hcgrCJUaJhcgrh-E hGhiw hGhhN!jhGhiwUjhN!hpEHU(jf@ hpCJUVaJnHo(tHOQ(QQRR[S\SdSSSGTT4UU_VVW!X$a$gdH$a$gd7.$a$gdf&dPWD `gdggd? &dPgd?gdp`gdG$a$gdG $h`ha$gd=VQWQXQYQRRRRRRRRRRRR[S\ScSdS{SSSSSSSSSSSSSSSSԹwjf^h'h'6hgjh?h?EHU(j@ h?CJUVaJnHo(tHjh?Uh?h1^hGhRj6 hGhRj hphp hphRj hp5jhGhUjhGhU hGhjhGhUjhGhUj4!? hGhUV#SSSSSTTTT)T3TLTMTNTaTbTcTdTgTTTTTTTTTTTTTTTTTTUUUsU~UUUwlɅjh?h?Uj4!? h?h?UV h?h?jh?h?Uh7.h-jh?hgEHU(j%@ hgCJUVaJnHo(tHjhgUh?jhGh?Uj4!? hGh?UV hGh?jhGh?Uhgh'*UUUUUUUUUUUUUUUUUUUUUUV VVV"V#V$V%V'V;VW?WQWRWSWTWVWcWjWmWnWoWqWuWwWWWWWWWWWWWWWXX XXXXXXXX X)X/X0X4X5X:X;X=X>XIXLXNXbX̶h'OJQJjh?h'Uj/h?h'Uj4!? h?h'UV h?h'jh?h'UhHh-h7.hgh'DbXcXdXvXwXxXyX{XXXXXXXXXXXYYYYYY&Y(Y,Y>YBYCYLYRYUYWYZY_YaYeYjYqYYYYYYYYYYYYYYY ZZZZZZZZZܺŲܟͮjh?h'UhHh-h'OJQJjh?h'Uhgh7.hfjh?h'Uj4!? h?h'UV h?h'jh?h'Uh'>!XXDYYgZhZiZjZlZZZ7\]]]]]^^$a$gdC$a$gdtW $`a$gd/ $`a$gdG $h`ha$gdI$a$gdG$a$gdoO$a$gdj3$&dPa$gdHgd'Z Z2Z3Z4Z5Z6Z9ZDZFZIZQZRZSZTZVZaZbZfZgZhZjZkZlZtZxZZZZZZZZZZZZZ[\[_[l[m[[}vv hGhhDJhDJ6hDJ hGh[}F hGhRjhoO hGh`3 hGh$hf hoOh`3hjhaUhj3h7.hHhgh'jh?h'Uj4!? h?h'UV h?h'jh?h'U+[[[[[[\\\\7\o\p\\\\\\\ ] ] ] ] ]!]3]4]5]6]\]x]~]]]]]]]]]]]]]]ʿʴʩ|u hChC hChI|A hGhhThhTjh CUhlh{M hGh h-Ej hGh{MUj hGh{MUjhGh{MUj4!? hGh{MUVjhGh{MU hGh{Mhuh! hGh[}F hGh8,]]]]]]]^^^^^^<^X^^^^^^^^^^^^^^_ ___O_P_Q_S_X_Y_Z_\_ ```3`4`F`꾷}vrvrvrhe hGh\d hGh8j3hGhR>Uj4!? hGhR>UV hGhR>jhGhR>UhGh][W6 hGh][WhI|AhtWjJhGhtWUjahGhtWUj4!? hGhtWUV hGhtWjhGhtWUhC,F`G`H`I`K`L```a%a-a.a4a5a6a@aaaaaaaaaa b bMbRbXbYbZb[bbbbbbbbbbbbbbbbbbb cc)c*ctcԹԹԲ۫ԹԲ۫ԑȑhF4j#hGhR>Uj"hGhR>U hGhR> hGh\d hGh8heh][W6hehGh][W6 hGh][WjhGhR>Uj hGhR>Uj4!? hGhR>UV7^Ubtcucwc{c}cccdddddd!d"dddhh $h`ha$gd2o$a$gdG$a$gdj3$a$gd*$a$gdG $`a$gd6L $`a$gdG $h`ha$gd6tcucvcwc{c|c}ccccccccccccccdddddddd d!d"d'd(dIdMd_dcdwd|ddddӦȄ|uqqh* h*hzh*hv_6 h*hv_ hGhr8jKh}<Uh}<jJmh}<Uh#h^<hGh}6 hGh}h I h*h} hGhhThRj hGhRj hGht> hGh8ujKhUhj%hUh6L)ddddeeee3e=eBeZe_e{e|eeeeeeee0f7f8f9fKfLfMfNfSfTfVf^ffffff g g g g6gvgwggٓو}jhGh5UjhGhUjhGhU hGh5jhGhUj(uY? hGhUVhhl~h6hl~j hGhUj4!? hGhUVjhGhU hGh hGhz.gggggggggg#h$h6h7h8h9hhhhhhhhhhhhhhhhhhμΡΡ·xtpibi[ShGh86 hGh8 h2oh2o h2oh8hhThrjhm=U hGhhjhGh5UjhGh5Uj4!? hGh5UVjhGh5EHU#j@ hGh5UVnHtH hGh5jhGh5UjhGhHTEEHU#j@ hGh5UVnHtH hhhiii-j6m7m9m=m?mCmDmmmmmmmn$a$gdVE $`a$gdG$a$gdX$a$gdX$a$gdG $4$`a$gd.;+ $h`ha$gde $`a$gdGhhhiiiiiiiiiiiiiiiij*j,jjjjjjjjjjjjjjjLkMkOkιեΌΌvrjhGhnH6h =jhGhnHUjhGhnHUj(uY? hGhnHUV hGhDh.;+ hGhjjhGhVEEHU(j}@ hVECJUVaJnHo(tH hGhnHjhGhnHU hGhyC hGhH hGh.h<% hGh8&OkWkYk\k]k^k_kkkkkkkkkkkkkkkkkkkllllll%l&l8l9l:l;lFlGlYlZlռܩ됅wljhGhcUj(uY? hGhcUVjhGhcUj4!? hGhcUVjhGhcUhGhc6jRhGh_Uj(uY? hGh_UVjhGh_U hGh_h_hVE hGhD+ hGhc hGhnH hGh87(Zl[l\lglhlzl{l|l}llllll6m7m8m9m=m>m?mBmCmDmImJmTmYmmmmmmm»·°~wlw^j(uY? hGhXUVjhGhXU hGhXhVEhY6 hVEhY hGhhTh cj:hXUhXh IjjhXU hGh chVE hGhD hGhnHjhGhcUj(uY? hGhcUV hGhcjhGhcUjhGhcU!mmmmmmmmmmmmmnnnnn n2n3n4n5nMnQnbnmnnnnnžyuujcUj4!? hGh3&UV hGh3&jhGh3&Uh3&h3&h3&6jxhGhmUjwhGhmUj4!? hGhmUVjhGhmU hGhm hVEhVE hVEhD hVEh87 hVEhmhh c hVEhY hGhXjhGhXUjLuhGhXUmnnnnSoop qqqrQruuuBvCvv(ww &dPgdGgdG$^`a$gdG$a$gdG $4$`a$gdm $h`ha$gdY$a$gd3&gd3& &dPgd3&nnnnnSoTofogohoiomoooooooooooop p2p3p4p5p@pXp^ptpupzpppppppppppppppp޷޷Ћ޷jIhGh3&UjhGh3&Uhx hGh wjhGh3&Uj~hGh3&Uj4!? hGh3&UVj|hGh3&Uj(uY? hGh3&UV hGh3&h3&jhGh3&UjzhGh3&U1pppqqqqqqqLqMq_q`qaqbqmqqqqqqqqqqqqqqqrr!r"r#r$r5rBrNrOrPrVr{t hGhhm hGhjj7hGhEHU#j@ hGhUVnHtHjhGhnHU hGhnH hGhs h/\j@hGh3&Uj4!? hGh3&UVjhGh3&U hGh3&hx hGh wh3&h w*VrWrXrjrkrlrmrrrrrrrr s sssss(s)s;sssswl^SL hGhjhGh6VUj(uY? hGh6VUVjhGh6VUj(uY? hGhnHUVjhGhnHUjhGhN EHU#jR@ hGhN UVnHtHjhGh(XU hGh(X hGhnHjhGhEHU#j@ hGhUVnHtH hGh6VjhGh6VU hGhr"ssstttt,t-t.t/t:t;tMtNtOtPtZt[tmtntotptzt{tttttttttttuuuu#u$uȽ鲫ȇ鲫|ung hGh-# hGhbx hGhmjqhGh6VUjzhGhnHUjhGh6VUj(uY? hGh6VUV hGh6VjhGh6VUjhGhnHUj4!? hGhnHUVjhGhnHUhGhnH6 hGhnH hGhhGh6($u%uuuuuuuuuuuuuuuu vvvv)vAvBvCvlvmvvvvvvv(w)w;wήtijhGhxUj4!? hGhxUV hGhxjhGhxUhxh^0j*hGhnHU hGh hGhYuNjAhGhnHUj4!? hGhnHUVjhGhnHU hmhm hmhD hmhnHhnH hGhnH hGhD#;wwBwdwewwwxwywzw|wwwwwwwx x x xx.x4x9x:xQxVxWxXxjxkxlxmxxxxxxxxxxxxxx$y%y7y8y9y:y·¬Ш’Ш‡jshGhxUj|hGhxUjhGhxUhl'jãhGhxUj̡hGhxUj4!? hGhxUVhx hGhxjhGhxUj hGhxUj(uY? hGhxUV5wWxxyyyy0|~~7΀$&dPa$gd#$a$gd# &dPgd*S$a$gdK $`a$gdG $h`ha$gd&H $h`ha$gdKdgdG$a$gdG &dPgdxgdxgdG:yEy]ycyeyfygyhyyyyyyyyyzzzzzz${&{0|J|K|o|s||||||||}}}ʹ͛͢viahQ*h\B6jShGh_EHU#j?@ hGh_UVnHtHjhGhD MUhGhD M6 hGhD M hGhHjjhGhZUj4!? hGhZUVjhGhZU hGhZ hGh&hGhZEHhGh&EHh\hl'hx hGhx&}}}}"~4~5~i~j~|~}~~~~~~~~~~~~~~~~~$%789:STfghwlwajchKhUjzhKhUj4!? hKhUVjhKhU hKhK hKh hGh8^hGhD M6jhGhD MUj4!? hGhD MUVhKjϯhGhD MUj(uY? hGhD MUVjhGhD MU hGhQ*hQ* hGhD M&himqr&278JKLMOȀ΀π#)*;IKaңңңңңңңң҆{jhGh9U hGh9j:hGh*SUjChGh*SUh <jLhGh*SUj4!? hGh*SUV hGh*SjhGh*SUh*Sh hGhm=hGhm=6hm= hKhjhKhU0΀`a ȅɅхUV02$a$gdGgdX<$a$gdX< &dPgdX<$a$gdA gd@W$a$gdG $4$a$gdK $h`ha$gdKgd*S$&dPa$gd#Ł./01σЃ !3456VWijkltg\jhGh9UjbhGh9EHU#jc@ hGh9UVnHtHjhGh9Uj(uY? hGh9UVh-4(hGh96jhGh9Uj4!? hGh9UV hGhHhKjhGh9Uj1hGh-4(EHU#jd@ hGh9UVnHtH hGh9$lvw݄ބ߄ȅɅЅх &'9:;<TUVWijklüzseZzjBhGhX<Uj4!? hGhX<UV hGhX<jhGhX<UhX<hsjYhGhsUjphGhsUj4!? hGhsUVjhGhsU hGhs h@Wh@W h@Whs hGh# hGhvTh-4( hGhAe hGhgth@Wh96 hGh9"ÆΆֆ  Vdjlt$*01278;Ծ hGh! hGh2c hGh~Z hGhm h@Whmh' h@Wh@W hGhR=j'hpOUj0hGhX<Uj9hGhX<Uj4!? hGhX<UV hGhX<jhGhX<Uhv{hX<h.B0ňLj҈ֈ\]opqrsۉ܉ƹز|ug\jhGhEUj4!? hGhEUV hGh)ijhGhEUj(uY? hGhEUVjhGhEUhGhEEHhGh2cEH hGheUrjyhGh6LEHU#j@ hGh6LUVnHtHjhGh2cUhGhE6 hGh2c hGhE hGh"#$6789DEWXYZabtuvwŠɊʊ⻰ԥ⻚⌅zshjhGh2cU h)h) h)h2chJX hGh2c hGheUr hGhrjhGhEUj hGhEUjGhGhEUj(uY? hGhEUVj^hGhEUj4!? hGhEUV hGhEjhGhEUjhGhEU%ŠʊUV܋8׌`*֏׏gdGgd+BdgdGgdo &dPgdogd7Ie VD^gd)$VD^`a$gdG./014TUċ89KLMNRtuҌ֌Ծzoj?hGhoUj}hGhoUj(uY? hGhoUVjhGhoUj4!? hGhoUV hGhojhGhoUho hGhrjhGh2cU hGh2cjhGh2cUjhGh2cUj4!? hGh2cUV*#;A\_`astuvyٍߍ+,>?@AKciz{ǎȎڎێ܎ݎߎ ⷬⷡޝВ⷇jhGhoUjhGhoUhv{jhGhoUjhGhoUj4!? hGhoUVj-hGhoUj(uY? hGhoUVho hGhojhGhoUj6hGhoU5 &)VWijklvՏ֏׏ JK]¾}m`}}j hGh5Uj4!? hGh5UVjhGh5UhGh5 hGhN hGhd@EH\hGhE EH\hGh1p'EH\ h+Bh7Ieh7Ieh+B hpOEH\hv{jhGhoUj4!? hGhoUVjhGhoU hGhoho!]^_`ېܐ7WXY[\]opqr|ǒκΦΙre^ZVZ^O hGhbh?hj< hGhjhGh/\EHU(jj@ h/\CJUVaJnHo(tH hGhjhGhU h5hGhzG5 hj<5j8 hGh@5EHU&j@ hGh@5UVnHtHhGh5jhGh5Ujv hGh5Uj(uY? hGh5UV׏\n9Pqrtgd 6$n]n`a$gd\1$nh]n`ha$gd&H $4$a$gd? ^`gdj<$a$gdj< $h4$`ha$gdj< $h`ha$gdG$a$gdG $4$a$gdj<h`hgdj<h`hgd)ǒȒڒےܒݒ%&CDLMRVfhxy>?noՖ"$%ƺ}vvogchhhvH6 hGhvH hGhjhGhEHU#jYh@ hGhUVnHtH hGhjhGhU hj<6 hGh@hGh@6hj<hGhb6jhGhbEHU#j-@ hGhbUVnHtH hGhbjhGhbU&%9:LMNOPRVZ[mnopØǘИʽ涯wbUQMh?h`|\jh?h<2EHU(j̗@ h<2CJUVaJnHo(tHjh?h<2EHU(j@ h<2CJUVaJnHo(tH hGhjhGhUhj< hGh7Q hGhjhGh<2EHU(j@ h<2CJUVaJnHo(tH hGh?jhGh?UhGhvH6 hGhvHИ GHIJx~šÚĚŚɚFOPfqrstxy{ț̛µ򝕎߇xxxxhGhbR<6 hGhbR< hj<hbR< hGh 6jV!h 6Uh;hW:h@6hj<h;6hj<jnhGh;EHU#j@ hGh;UVnHtHjhGh;Uh' hGh@hGh;6 hGh; hGh/:Crsɜʜܜݜޜߜ`aKOS=µ~w~w hGh_h' hGhgN hGh&GUj:hGh CUj(uY? hGh CUVhj<h C6hj<j7hGh2EHU(j`@ h2CJUVaJnHo(tHjhGh CUhGh C6hW: hGh C hGhj<hbR<,ɜ0@ACDڠ۠f\ $`a$gdGdgdGgdG $h`ha$gd @ $`a$gdhT $`a$gdG $n4$]na$gdj< $h`ha$gd&H$a$gdG=>@ABCDIJLNv{|֠נڠSTfƿug\uuj NhGhpUj4!? hGhpUVjhGhpU hGhp hGh[ihGh36 hGh%"h]Oh36h]O hGh#hGh#6 hGh3 hGh_ hj<h_h' hj<h3hhT hGhp2jv<hhTU hGhVT hGh C hGh_$fghiݥޥ(*Ľ{phdh\hn h I/6hdh)h I/6jShGhynUj? hGh I/UVhynh)hGh I/6jQhGh I/Uj? hGh I/UVjhGh I/U hGh I/hGh I/EH hGh\ hGhpjhGhpUjOhGhpUj4!? hGhpUV!ɧ%&+,<=>?@Z[349:JKwpiaSj? hGh I/UVh)h I/6 hGh`R hGhx$}jYhGh`REHU#j@ hGhx$}UVnHtHjhGh I/Uhn h I/6 hGh I/jWhGhx$}EHU#j@ hGhx$}UVnHtHjUhGhx$}EHU#j>@ hGhx$}UVnHtHjhGhuU hGhuKLMWXhijkªĪŪƪǪZ[klmnЭѭd»»⭢ԗ␌shjchGh`RUj? hGh`RUVjhGh`RUh) hGh`RjahGh I/Uj_hGh I/Uj? hGh I/UV hGhr+ hGhHj]hGh I/Uj? hGh I/UV hGh I/jhGh I/Uj[hGh)U(@ôĴǴȴɴʴմִڴ%VXfgƻԷԧԠԜԘwk]wUh6LCJaJh6Lh6L6CJ]aJh6Lh6LCJ]aJh6Lh6LCJaJh6Lh?CJaJ hGhhhSh) hGhr+ hGhChGh I/EHh I/jRghGh I/Uj? hGh I/UV hGh I/jhGh I/UjiehGh I/Uj? hGh I/UVƴǴȴɴʴմִgWYQd$e^e`a$gd6L`gdG $h`ha$gd&HgdGdgdGg./CVWXY[ ˷̷޷߷ ̫tetj;ih6LhnHCJUaJ#j4!? h6LhnHCJUVaJjh6LhnHCJUaJh6LhnHCJaJh6LhCJaJh6Lh/_cCJaJh6Lha.CJaJh6Lh6LCJaJh6Lh I/CJaJhCh I/CJaJhCh MCJaJh6LCJaJhCh/_cCJaJ$ +\fxVWXYZ[ݹNOPQRֵ֪֪֪֟։ցֵ֟vkv`hChnHCJaJhCh+'CJaJhChMbCJaJh6LCJaJh6Lh+'CJaJh6LhjCJaJh6LhMbCJaJh6Lh#CJaJh6Lha.CJaJh6Lh6LCJaJh6Lh ?CJaJh6LhnHCJaJjh6LhnHCJUaJj$kh6LhnHCJUaJ!RSD[ JLfgɺ겧zrjhGCJaJhy-CJaJhfeCJaJhChfeCJH*aJhCh+'CJaJhChfe0J6CJaJhChfeCJaJh6LCJaJj mhChnHCJUaJ#j4!? hChnHCJUVaJjhChnHCJUaJhChnHCJaJhCha.CJaJ%dxy h]h&`#$ $ VDx^ a$gdB$a$gdy-$a$gdGϾwxy мwoodZTZTPZTZhRj hRj0JjhRj0JUh-hACJaJhv+CJaJhdCJaJhAhACJaJh +5hA5CJaJjhv+UmHnHuhACJaJhBCJaJhhTCJaJhKCJaJhdhK6CJaJh-hKCJaJh-hK5CJaJjhy-UmHnHuhChy-CJaJ h-hACJaJhnHh(hRj hRj0JjhRj0JUhLt0JmHnHu !"#$%&'()*+,-./01234556789:;<=>?@ABCDEFGHIJKLMNOPQRRSTUVWXYZ[\]^_`abcdefghijklmnoopqrstuvwxyz{|}~$a$gdy-&1h:p7(}/ =!"#$%nSJq.MSPNG  IHDRPLTE xx0((hh( P@@pXXpp@00`HH``8((pPP``ȠxxP88XXH00xPP``pppHHxxX88@((P00`88X00X((@880`XH@PHxH@h`ph`XxpXPxPHxH@p@8ȘРh`pH@Ƞ`80Ш`@8ȨаhXp`xhhXp``PxhpxȐxиxhXHP@xH8ȘȸаЈX@بp8@8Ȑ@Ș@РHبP @Xh(08 (0`xPh (@XpPpXpPhH`@Xxhx`pxXhP`xHXp@Ph8H`(8PPhH`X`pPXhHP`8@Phx`pXhP`HXx@Pp8Hh 0(8XXhP`HXhpx`hX`xPXpHPh@H`8@X08PhpX`PXxHPp@Hh8@`08XX`PXHPx@Hp8@h08`008``pXXh((0PP`HHXhh (@@PXXp88HPPh 00@HH`@@X((888PHHh 000HPPx88X(00P0(HH@P80@( 0 ( php`X`808XHX( (  ((  H8@8(00 (( @(08 (PpH`PpXp@ (PhH`XpPhx0@`8@XhP(0hx@P0@x(8P`X(0HXP``pHXXh@P8H8Hp08h(08@x08HPSH'JbKGDH cmPPJCmp0712HsIDATx^iTgADe@@x0hYT SFOb0Q8b;-4F l-0LV:9T׃ڻyd|5o׷ϟ?}]T|X}ojkZv6TTWtW𾺾b}QqQQa}}Ma1_)';P5E55%%%[[KK:[k:;JkJkba _,/)).ZS믫;/_΋5 %%%W?ŋΕ5E|xb^kj.[KJkՋ/+..Gp~SZ}>ϛ?|uagQ7իW/z|ꋪ;_7E|T@!#ޖ𯰸5% ,immmhhjjm8?P %VbC`u~;;A]Sů"n]ZuEMq2@zqB<*\^5m˥,L 'i1z x|ѷVg [(kL_Qa[cCikj%W\R$% zߺkC ?ZsQ-&Twlb(+hEjuIakM@^X&kcg Њ[5%Eߴ#6!lsABHUGBjJZj޶M[Zz)r`֎,%N4cWT+"x|J= + ,WS|V§" ?>`gN"zZ+ąNl.s (q_[Z^aw`3>ijo%%64j:AD H-P.s}!6&rđXdzE|<`țZԾWF(|6Q"1VF1?SrqQIteӟB_[QֆK WV7nWq? .^t!AQ!7ܐhBhq7qEK,PHa0g7]5oډ KzHy'DcjCTh5i ܢ,|xg}k||p3ix*4 7!Β]hbEĈg+16z")cf!@>Y(/HK%s0=4Pyʊ \Z]1oEL#)ʆ. Fno {< Ox-j|;o65aW xWb e6Ip?5D|RZ`6^,䍾$(8ƒ/!+QAtx:d nHJ`zqkGf@ɥ۷n7>H0"o<<gO= \<6ZXU[ms쾤LTj,:(  ]iņEP ck`Id Xl3zYDAk,$Y*GAB %D<g8gP[-I;>=AA4 掦U]Y 6"% /6H_t 7x/d y$x8 WK_D-ÍJ  PB)|(r|.3|<|nW8;<[؀s#!(j(ZŘj/j悊O-QlTM j]դST,~]|XKJ  N+H5me(XmB$mʇs>XD|>3789y=ÞZPrΕƆSJ4Z꯴]B~P+eWJ0ls?8"ۃDN̳bKj+eLW0؄dM cLh,m@e;z?;>9ލ 7 Ղ660JER(ӕR\0[jTӤ(YKU+P< _b)-BY!Uȱo@y챡v3SSDD 6 7d*zA!ArU38 t0D`jm+kEjAY 2!eu}Kg@фοR8Ţ ]mD6̠ @ɚU4521ÝqǨbH}'Լp>зFx7/ =3;8'Z4\^un JdmŢ2LGM/r{PWgΣji}=82jkW{ick Jp`۽Crn%boH!Ļ$eSS^^6%A LG 75a`r$ PBֶ0n]\9>}z{|VQULRѓ] ]JW#Ԙv]ikYc[~RcHTol?6}nɿio$@lBvY / y/JEp?&Ky 3Vzמ>mwceRDaZ !5H㻴d7 |'aw$\09Ae 7F H}`%8t"M%>*0żTU"&v(1KH?Ux|V={nB43ZtkV!nlx$bTȮX~e TҞ[wtiR)"ϸ&!ss{ܺWeTuOB~֞5=J_)d]s#?^u}vAB4Um*ϟZU!8[7$ȼ9l]e~~ |']}= \iBMYQw'q-B0*QuaΫǫgR[%Y'l IK?Eg]W^nUWh,"RT@i@JM<'Ő֮ \tG XyF)2d8Uk}\.ka'B LB@啄hzowxRii:+_^b bP"[%%R0!"CCLt k׳U_JxUX 9-ca }H̴AF-%B%F$SB)V$ q/IS5DsqNIYL wd^V5<5!J>88Pk^.-Ų,Cٱ_gحIB+Pػ?xڵABw802s9$5'&v̋-L{{?  "<2oe9$vYukVdyS^}}{Nw ]R`\cC{((P*p`6_;ܯ1g|׼?=g) `]UFUoLGB'a2`IA Ha_EˌGmO4. ͚uڻgUVV$TY&f]]q r[*Zv?F3ɗ^CL0n^@ۊUrR"o'ExYÑx$esC[+d&10 N5( ۨ0W_ԠxRSeeΕ8{lBxzUlAkT6X@usBnCW\t~E 4Uj˲B8p*Jb!kω$e׶lAF6D2 #,iWAkCYcEө:BX:^ 7(7\y솫P]]_A4'&<|tϼ,/DC*䘦/(1gO)saTղ酩S_ؗՒ Sʺ#W_iyQa(Nm;YW*aP ̞`Ccc`]QtȬQDZOa̰0L߸NgJ&,1eS`gj UCxPO)ʲ$lT "Ozςӻoo۫8J+N55-;85Q*?{,Z(@WA`wi.B5r䁕Y1J+Zu,(PJϞQ-3x1} 9nJdR|A=O}P>KP'}Gg&fi3+*ζee2}JϕQ2^D}V VC͂4Etזd{Ar1VJזS)U6h43cyMw5ht0rWa ZW~dSؙio$\}}2Hkq>RT^^f{;¼9j'!э4vGQ`Bأf0mrnL|)"{{*a^.EJ B()/-7Kɾ<=RO Ѵk1>=yfVķT0fWV^+$f\g1Pc%P$hAȜj2D3D6!ޑV^/='&AMhPHr|SNzDT#Wkv=9=:$Lt=r$Lx<*JA$l]PQjoAKJDH$W7F98e=l9bG}|,GD}>U I@\/< u*y?q:CfUF|(,v͸Mvfw= {'(;E Dt<\y.:'%o\3z!h%7;:3sv8cC,5 \(49ه;H0;CKT]yY M핕Lc5sCP:9TL=VL?mJ=TeH\N/BzK z0~>{/&ooK%˔w/kT5ebƻor0gP>=ʴLU5T1֞Ժ+!Gޡ~n_f&_<ЗY/20ع K"@)Ϳt25 r2>AQZ:թDZx|KB,uS{Dab$,dX>t+* l$D53p1}m+Y쓶>o8McE}Q/<ُdOgZnF+heJs0sZETM;%\g#Pyf="RInLrq}~y;"Jr@քow1\AnVrx?LʵMCȳ4X,qU]oAPAYS^&O>yr,"D9\+%R{z@$A,DkV q*0R W +w,7ZD/ ';ESM01nrx']@Iw]3>9t~SsܮjjIo)2`NBcrIpOjNS~pavOcETDו%8_>~̝ 0sw]ӓzhg)D##|#C|""L N DX_.u &wa,.֤5eW<^Gi!{N#_iMejKFUQQ~x k`( 'AX51 '::x,Ӥ)滳x>]{kQli\Qr9KD>caȀ`}aPu%uh?$BZ\9)",_>[)1Snh!;-= h fwX)={8DEIPf=358˽/!19lĬЀ<$y&DVd%kDֈ_2a Cׄ٭l$K ֣([З*| p¤zЋdNTf\uՀ$!*qLјly[!ZN_xI)JۂɁ`iP psĄI))RRQ#dQyy~9Sjff(!ZUyCaX ~x-rHfx#wdV1Tsr[JJX)Wp}#<!I6Q@4\!g5] ~A#5н>)))&5=54!\ (⌊$H^N^^4(b!8H=9UTs!pQ>H(=AeURo=x|z޽jv4~l;*\wf5Hs-*N"&'$Ĭ$eIL !$H 0?B/^D% M2%BP.EjBf#"A!r7D'^HrG&%`;)#%+01:'$yCfHPMK ^Q/- (h0@j- ˣ\)k9@LYY.O#K0 ɀ"0@f)CA,_e͞<"}Z|w"})f+{;.StQ,Y^|*F}PcORSu`aǂRNI O&QD~'='54=Xr(SCC`!-T8637"7!KqBq#q VlP+˧brkw?vM(=M$ a(H＀Jh21ϣ@vxdZZJ8(2ȟ#/%58"Jn0B\H/}_Bq#iC")Hg!@Y" YċP! M"/)z݃U#g]2΁r/w\FIEUcxG>e!fX Ąׁ 4,,,M@ LN&U LO MtN  L2o|E^M@a[SN*74Dϗ.70_YV\V맡jh#Nܪz(*5X$%>Vp𴴴lJRZF܆)ɉɤ&KK љɁ ɡE9H0_?p@yAy~0#J dg?T *KϢ g:% d${-:#-Hx0،,X-6)-=qyⷨ5f^ݜI0h/a|0 z=d2!*OKKikޘjoݒIDHRB|k%<=rAWO*&ČeOAW-9M<@nM" -%P񦐱Qc &+xKN[驁fA@xhaa<\~"OI4tT9e-j^J%4(ĽUti Des#T+G\/uEM \B%E]츌H37#8HzI!+ 26v˱tP$$ %ɝp 8q݁$}&&R9%&à, t' JPQ<ҭ3QKn| ,- dO"?fn*)v{SRY$5'2 k/ [8 s=ׯMq@#SM2C>eLy_$'yir=%65`X`.iiqS&E#`1*D-pA~Rؔxd ƌxْl!غu֍i]?sgꬲAJQ,Ib!GaJ(i) qC 嶏rwG[` @ L$/ T7eTI(CGh޺۶7Ǝ Mq88o.?~).{0 #u0|XFosTgXM C 0)xėwG۱ʱ!W)qx{ ecSGp̜_EC)!a=]邫eFZre*rx]o+|{{J;bnSrT^/#|,Jn Q#02yp0pUT &\V1Sk0=Z~6)ܝЃUP[N Q* S Ⓥ 3uND5c)%޶{wEPZ}s 'W2g_aMK[JHʣT‚Q0X\~JZ bNH @8ZPT)T<gҷifg=,BCUjCXFRzp'#iSRPFEe&g4|2Di!Bg]w Qlۦ0+ysښ{XWl%nn*dj00 Y .Pf^QgBa/PИ>("@FNi~^[{yiLUk#\RX_jTԹGIpy …9@w/{)˴FT@Xy8V!, 1MH(DxМ_u>k=D9<14 #\2Cb B#W38%iwV4 `F3XXLMϖAP䗃Jr֭(\ Bd"􀤰 AP&UD,$$f2Z+^yZۨyv©FQ2o?}wPS%Lk4*2ra_D-[O (/Aj FDHjVa<b_ DܶX>!"AIc=qfC@eߡ 96[VU{EL1IxV$I#z'lb{{ᱷ0HKJ@Iq]'J# ;SUJߑx=vqܕBB=܎ĉ@A&c&BE((Ҥ#IRou u (1>Ik'-E 'y:~XF} ST?RYlӂ-zK L5I7nj Ь@HP(jgP"ܹϊ4,Rc?b씞Y,/j|pAtGXJꬨH^mCP": Lٛ{J.2Rω' Bc }iVE%.;ȿUYLE*yʱTJͬOҩǜӊLΒ׃.9͸gYB},4/n]2 ?g֍p%!yP:fdZEZ$H] g(!$ rG0ZhOe3)IL2 a܌j<6}@V($%w kE*m(9ņJD0ˆKj y|G|ib;0pjO &di W*S@) nAB*yjw\\~1:#=Ig5C3416+5̝:(yw$>2N&3Z4 ՟o йAVWIHMpN'tVV`@&O% Dt} (DǛ?р[?|̜q-M[UT5'ݰiK02rɞ*J*LT K.E15U<} 6fGbZMr b3$xKSq۽"` Pf-IJ)„֊"m(ha;B#ݺ) +Ť51a!)l GO&ݣ+nfCMc '*X3Z_K{ț?Mm7#Z.38-ba>$~+DVWjF=:RnfzRUф\pG= FPs=d ˆeBuFiMC8sC0+#KMd+\ZF̓#2}{HPZ> ul2V b0iiwgCA  ]P 16:nKs;ak%'sFe;$(VJ%D[ "ѾlH56֬3+)j5VVV,( c m{;-h;Np;('#=11g3veHЖ Y~Ks?AYkfT۵LGlSycqI $'@$!BZ޶O)"6]̂Զ"~1}wc|v8KlxqVCeOrIPpC֖&&e'-;ԚͅYl0v%۷zhR@ݎ6O fА#uf"!*H~ =&0pl/@3)uЙ]$\z>|H˴p=kXSsK,LKG[O|ĉCq{ܻw ظUemC 8.'_ ͣx45s9kVh uK,tIjqmUUR??:(E_i %~KZ#92BEAHS6`Ksb0Γ&JA4[E'ጼ*;=NZJ)XMڰ6+4爜c[~oVi}Ui@h|2+e2?/٢V[W)rk E} p`2t/f7D_z}xab;*Mc넵I oHnU_%}Dqz̥q8* кRZ0e3FHI1!-4Y:ov.vU#CQv^/Ec2%tiiȝt{8h%U^Ky`r&;Kfo!C @E} q}g8@G9U'1"AϑڎtJR#Y !O \^ڽ.XT63PB9a}aģ͑शeD~@/DQ^t &K۠_S:zT$ctHS&O=I[wRRxqMF QgơC`%@R3Gs `O[o\UX0039HN)pFJpԁa!/}kmXu(;zm0CM\Mf$Z\tֱK)>^$l*WAx=6m} +R&@yHoXdƾDμa[,{z-tzgIRèeNYvkHse96 +ű4Aa"8U! B A6 wԣ4]r=3r)87iOq*"uwM'A>UiS8aB[4_i ۨ}իe$Ǿx-[b!KRI=c6+ڹ3|Z# ˙%jx(,ce1{ צY2$Q"k&,okmFaMp[LILÎoz͎0=+<%RCUJM.(S;ZA! 0u m-ɴ'ܬ_ >p438jDM@#"V~֮lgA Z=ϡpؘYVOz+#0̰KG9hdIaK]ZTvP+wuǩ߻7?]Uͣ_Nq6J`Y1MXx\* K} q;[ᔁbǒ`L">"ӭ 9Q/i,cstUqVŸJ,ԅC쩺;^NZy^l8̶Y9q%9ɊǟD|m>U%m$2N _ ahG%cuy]TWƦ2N,Sܩb:ͤBx֟Ţat71qUђiFXj?f#ÉmfC|z)HWZir=ڛ/FT%X 4kr!qUM8 NЭ9q8oVu7ޭ<6s>aK-e/58M231j~TGccS2A@0au(&DNDO,bQ4 |Y lvlFP;jc.A!#Mb LSՁgke!4yRn?tx][S5<rMu2)LiC@!޲tM=c8KkHdJd\( uPWƗ*.aȵ`DtrFS7@/ fNWlvauHNikkKPǟMQJ }_r=qPeݘ{0?YT{-[U6)e9ׄd O'ry~R "mF,"n kDz0 %~ϛz!VV"B$eTi "+7}s[uSS %!oG>-h9ň=J _eDwR2#GS4OFچ[ ʸn3B`eXO&RMSCt;gwsdV;5J/9x!Z=zӱڴǧūKrbj3䋋:H+'3ojta51 K/}ilm$TĶMإQj6[(isCU^J|Jَ>XZ\X XzYTWx"MEnriIRyaMMHQ$pZKL' ߏ>g&R])"gUc* $ ݒtida0^}C(kU"JZ']܌Uv43C-n IqfFH;s{na(dT,#aݞ8^Ke~?DIQj߿'`;=5).^Mh]VDlSWB+Ų$Ǐ!n$t99 rM+'7lm) =\0=l)$nŝ@G4t6(JJ&pZT-۰7 Drq5t`gJZoA͓]zS(ITa z8:ga C@،pWNg^&;i1|\&2s̈ߛ>RMev`rHKxcAgc{\(Afh+qe㪓CDt [@*EUXN!dd&W2;^ rP 2 l2|@b5'4$^?|ȡ]in.b$Hղ$)9T֝,IjI "a/RY`?4'4 +v$7;?_Eib8;N_I@t:5:vR}C7ٌû3hvzrl~^" LGC(b i Yu琅{"fu?.ԍ$(.,Y*SIZ-f.aQm(GnZPA> ŖH?kIW}=YlĪr jbt{x`ڶms:ڵ 1xP:4*m%Ґ׊/uϵ6s6H4?Ғ'$]?1 x[d f^_FW\'wC5JqAα7,e2.zj $F\'L8f }%%٨9;d{BJLLf\^玌,@Z݄ Au'fn&kI05d2ּl!6 !>" )`Vl]Nԛf6iS*hJ'r1YV3zSam;?g:+ӄkXO@5l֒uIR(fħ싥aĪg2KabHLXeiQ^+r8֊"eyK.$ a˾n<۵\n:<ܲ zښ ,E ,K'dDYf&CR(SU0 bhɀ/'6^tQjC`>Tuu?0WȇڏACAT~Yka͕DNIf%pLͦuM;EIpz[i,І=sR ;C Pnr0V83+7mSz-M^d =='ъ QgO f KW5 abucv )5o`B`jH8R\b"ć ʫPkIԬ;ti)Q J)+9'8( ֻ U0QK(5ngߚE ^'YsQ>^?zx)H VHHbXfdJ>p EdyOJ@l4Dh=.~hխeG9zpW!qji8㝂n.Q`ӈF 䚸9bw}(Mh3ljv9xekM3zg 0BZ4հO"V/ m8 HӅ\.Prb|$Yw݌nv`7;b $I&6|xPD'% $dr44Ҫ\5νLܭlp۳EN dI$IJ#s2y `ŽڠX7E;@@֡%M0=o $VfЅrICi泬4-]cR؅IeU;$rbrT( L[SAyރ{v8p [#yGiT8nOLuեIk`0e￿Ze!/*v5_PL~g^fi\.v & Ў("bKKb{V3y2^^„`BA)vOġѳ1{uVuxʾ-,FlpiSf#ҪM<p;2-lg8F&Ls!d6jr!n\a[bcl/3:iM:_"G<+{`)r_2J-DVQGRaOg.@+4>FI͹!aj`ZI Gߕ5(IBshY L(?h ۶.&ڛa}cX w` 1ifHLb!^(-&E@f4X'st ai}-lW]",d61T|$J%#I3x'YN:[[;:K KP'b˲"$dÓCٟ~9i^υk}lIF/JvhIQؑA :1VW ΪiI@*a֐& ( MdzrScw}rl6, ]4F{ZF|G ^D驂,0\^@Gi)C )ec;hESTX,Z"ta» UCӼ bTM7ψ#Xeu1 %Fq9<mEVN8xDyfIHdZSBd(eASu s2xOz^MʴSBM@ˈc{vgGPC<!it*GC\lJ^pb,D535FчŅa&'StUݶ5~E$,Mv6"0=W2Վ8a~pr f3܉*HW \D6g%o  2_u4WsfԘjt:˱ /dM=ܚ@$iw1Ԙ3g㑩0QYSI]״ (09k)hO&.9%t_m;$;Ra%e`=7$1,STQ{ܓ*% ;:o JVCGYhıeȽV6*a[gJr-nܦM{}5Mε:{X`"a' k͞mX$JGcL=. 7\0,e)koD|!J%R q(K÷徝,J^,,e M^8\R#RUWkalU qV${ x,[E{SUjFD)"0 Zt _PuI|R5vON+pfei/p]9I|׵kzZOP.YE1TIS;IItSuC89Vu@hdo-L6K !XFg= $JWڬriFQVQ[Y;}4N'uU!Ty _.'1  M.UNؒA) +Acqc;@HK`&MYhJ G'’T eWUTV=a`KBqCpW( W opk-wV;Qp@ϟpכ92($?2 L5tP (4e6C8J xVV^+KQkVS|>g!Sa5!8SNI;K!;qʼh%=l*1kEa%kh:;"!)^R :^|:׉5<P)֭c4֚V0D I_!jQZ ii3Mҩ[J8IKI1sJLҢ7{"l-TwV)56(aybWy?E*|AΝ'lm~O2ӟPRV2_>:HGU>'Z8}/jS㢶^l!!&VD@Tbv6}x+~ȏEwX@/An&X0|'n(v4gg\x= #7ɁpP=ArJp;,D];2| 7{MP5rPjɞC~'A{  ܣdj67xvQ8I38FF Fb,o"(-I:# $%f$e҇mtp[7y DUwnc4v|ț*t\(YµZ,OugmL}0^= a@;Mwb~k2ѽ!,S=_v`怒8l^k'b?1V=f^@W!V>X.*&e.4.\:-+q|PV]Ȩ(WhY"dJK54 jOpcƮ3Z@#7+V[9NiuC.5#ȨKC ?` k D\n3_58x64ҩTA:fp0F R?y^^P)eb97(tǦfee y}ڗy<*WѤTAQvdݖ:ڥȗGfȵ*P ''ZZzIfڽɵ d*-3-sZ344&SN(7-3vC<$YHE(ͬ? p: 9r>/LܤøF{}/8K#y"RԼv! AW^v5ISP͈!>;2%&&r3 rw&SaFӁj#Y߰D7F k]⩭f)0rcU(,wwJi %{tRqdy3.(`##D ,y!0=z-Y>iD[u V*(?N,]q),m|+`i)N񶤘u蠿4aGJ_ Zk) xeh˵*v!}6N[)yb@$OLBP ޞ`#, &~i Q(~=*']BbA+x/?x{K(mĚ YJBgYp![ٵ-m'{ZMMd 9k X=3(&ْuVk$نB2H[PR]y{iLpefgvk2uSv=bC{kQ0ce.M3n QSt(56ʞx8]GmfGvr'vNXa6*CX7홿cbnr +$KDU!>htPuh qNGGG5ijJoR O}~zo؞I,A)ή^d w`w+@#9WHN4WELYUD!wBUg\q}Ra 1Ykebd4VfjE-Z7P [nsw)xnlAɋ@Rh-:)Z'}qu?誜Ϳ;YhA$dW )&;(*L ޛ$zZv{VRZ6~7?Ddʏ!J$Ws I] XEi[i~l  bj/$ى@C)nLrފw@)Q+:92]/Is2Ȗ GADJndU+@EևzSַ?|7wG}O說!J7K(h+0qX&J^e7gC6C?'Ii;R) 8m33/Jl3\gO-Ye`ˠt.Ne> JFF(>|C &g? 8ndP<| k Ri4̺dnk^4\V b+}p5!dr :zC_؛97HѢG(P ޽3.( 拓T׿=Nfp @Gy}%籃/cMZBMji- )8 4ʍ̺OH([ޘH(NUO[##I<z:hqi_ ;.Tl`Y@d4]{{8A>j;vk+QM s~(w)|۟S0[\6U2Dښ6h%/z\eJ7V"==&s2qzkS5}˱.Nf \2qqOy u10Fq N#1E \/Y:;ʷM5xM w`yg>=5 Fi\TO\m5Pm'?e{k `X^Ĩڿh8vSIRLf׸,ohlfXld_v'H3XU@GZbRVW Vݛ0^\4,ɢ Go4+_B\Gvf@й׉F%LQ`=L @V !5@KKPh87#(*YA}t cB16G b݈!"iz ; k %2$ _x-;1J#sk ԺdUJk9LA˔jwxFK<]/4 ޺Ê,i_=%M!:e/!{Bz8|4EI=eUw$g%Εᷟ]}qMYH %r2)TVGIiE B@1Lg%v-Ճӂ2n/Zn۽2Vt0Kz).l렼wʖ+噶Z0!˂,ϪBz=Ħӕo !w\>i$--E5|Fت&5Ԉ$5`־3X.:?MX*]q g=&8@6)tdNt|,`˃:+X'*+lOW!7e@?lwEM"@Cx_:A eMTH^t{Җme9M? "LP~"M㧟~mKRTe͊i"NILDi&FI\0v&K5 ++Ĺ#U4!nJY+aˍdS }CxG+oԯêhiPC[.Kue ,go<gط?=rd*0xܮMh7gmKgS,i29&z6K3f7e0ڿ?ZQ&(Y,.2f9PD *=QUPhm#sf!LNC.镾" <~GH>}ݧt,5 6y!AFݙ~ͫؗMj6礪b|JOۯRB_? | MH#xDi8dN,Y9#LT[ST% &،Z8l gKiZڭ˕]JL~Mh^(o'E}(=I eL[I3PRHطw49Y$?au4Й_@QXXm#Fqݖ_!H=TtJ?XnȺ nGvx1?D\N\yӒjhP"Gq7#Ӡ+;tƴg2jJv]-FX_CeR?'̘u&^B&KZՃ(hEABBBiD`.'>a@{߼ #X$٤ѲOgIA.J$ƬEF:>=\ LMcXHqi.h;(DetLγ/sp+IHi-9!vHmdUU6B‘߉>;.KH#<:1aJIY9@HNei+*(}R5C"q"LY[CQY-LB OT9Α^k~MZ #}]wɪ'kvA_p&Wiu0@qdf 7F ){@ej_*v@cvgj2k}j{S(b4w#@%%.ֹi5+x3.r 墖_.{=FUa W-_I BoP)~dotuƩlzW!5η ' D ܡm?Y.3D'E#+_- m;kl!PRq#L$8"=)F92.>b ?#ЯtEՂQ|+70uj9l䕿ґnّ],gN66%nHhx)A2Ʋ—"mFR.6ކR.bX3{_J(%|𕣳RV<|UlIڗW@O.h W`pAVm:Ȇ%{WFtII5>.Z\Zfo(r6z4GZ`}a<_JO[d u_YJ 3wLГ^ /i+4Q IsGL,ED+Li{-vhILcWb`E ˮV"rh1T# L-n8l.37j'$%Wr} +V=fق^ת>L@/ptӏ?"Lګ"H<+b~18ŲIV9ݠd"Y(i`$W (2B D:DkCzW s)aSFqĪei '_- ~u`9b€z?@vaTs+n~>XI(rJN:;*UήYJR5fu)Gi870EHD,/Ni>>0M-W]UtFb{=N-r7}R$/9Dȑ#XB$Dhqo -`!\eearLLdLGfLkzs3>.)#ת@" q~EBkkߩwOǷ.*`pӿC4~Owれi#F@ DMaL~[a+"9SU50ilBsݶ@QǶhf{~ѿzw"lJN.ah͇^3'u2/F\ߗsdsa )Í!½EQc-TBE ,9`9>G[')rhv%k A܃~>+JV>Wj/#(,=pGgc>^Mʷ?Aaa5M\kAp/sHq:j8Yi@ݖO"-\˶Wu1K{PR Ҕ4?+bs% Ϝ, 5@k?>#H7xRg}3]dTceѨm+&t0e(XaaZCa"H5KMUQVRg6- t,)7ka-G1ںjbW}p f?vj@~u@Agɨ}}vq1a:Ϻu,k.j]*$TJ]喷F]ˢ\3Wr>&Fx#ek4coAȏ|w_r$oԥe˼|Cm-;؍t6X Rq=jj_O9 :UY`<+8Kccٜ/:0GŞvB<jg4p6e)*) 6Bg^pg0'A^;0V 2}_V:50S+X]uA{UD%/ QY 8qv5\IY} @,ګ%~)O_#WX~ ' .Qa6lPsK&िGohrQUk[R ~xTLI-C'JKEO6S6Ec#kQ š2JHz(4F~~s|a^1`&q֎ a 'La5T+83SRc fkA2IX,TiMψ , \_8#L*TxpUBiUj~j};ć&r@~ˉae#TrX5$/lNx9Y^fX;;PuN0*6w(+I t>K%-e59p&ē&4<3əB&p@hٳy;mi@MBi)0-|^Z̜{NEJpEzevĶh 0rH۟x7GFmTy ޡTV%*=i{/^MoiP~vdݖ,LNNNm9Xף8}ԣlGKJ2}"uKuRO޲I~i7=V"9X˵k)prx2;#oݿF0[*`6VY>Hǁh{Ʀpַ󤵉GQ[h`9χx$s8THxo'|{=(nӹ]jd s;n%_ wݑqb 7V< g0E\`;gMaҟ 8RAfkoI]>Do\wO=4.WI:C!V,LQheviVt$)LQhSSbS֝V"±hPl¨(_V^rK5鷿B=zW{KGh5XUmě4&_NPQ:"U:RD $ͤ9Ct5^%6tiٙe׎Ӷ& l6cs^fAM_؅Evd!P~!]FDiB~yK|Co j"̑O) v}^}pzu,_ec2 f`09%K͸&dɦ76WOm3V$R tNzXV:Cf2q|}3KSYoz} xR^I"x-UNkA 00Z*_V$hUvtgy-<]JN(,gE2LVy/'ufUio{~T*BT da%"-]&~(qH#et宏F<Z z-i.@ڊ@G"VVtf0$Ct9Ać~_?]85; MP/*v%TF<#ovNͥ%/TBHkO^A $LjZLg*j89SJ8ط~bA,A ѯ|9/|Єޞvp*(#1Q rijG B #°G+f*;"iɥ!%L=,36!g2NenٲY#o-SA LJ^{ 5ҙ4;Lt:#cYH[>^rcU ebjG*\=h?Ȟ9Z&=!M]2DPWE(H݊ˀr(oe.+W>}31sob}73+P-Pxx׫[9]˕?tɲsE˅M^5yZ9ԖF]wdZQzۨ^:}L*re *?/ sOeH;=q$+>TMB#Snw}ߡϊвo9WӸv*A④>¡$4ID_/΂vHX,@zeytRY9[Gfwiʉz3!Iv $6֐$3}畣U 񮍜nBҜĤ66+(0>9c,,c% "{f}wKV.s(O+8x{QM@Eqij-p}^:4m*IENDB`F)5e ogZ>(JFIF``C     C   -" }!1AQa"q2#BR$3br %&'()*456789:CDEFGHIJSTUVWXYZcdefghijstuvwxyz w!1AQaq"2B #3Rbr $4%&'()*56789:CDEFGHIJSTUVWXYZcdefghijstuvwxyz ?V~5_jQUk TUZ(6 UVQ@SUUV˔G(Š((UZ@Q@Q@Uǘ( VkSY\*V5ari<<R_j^Frm}?╷퍭&+䊻=սͿ+崱s2\U}Eq_ ym 9Wk_֥*59.&2k-]Q\-E(tPwU*Tjk+EW%Nbz ޥq.ի^Gx=tK~^ ߡgϨ[zAfXn:;K:_Ƿ{Wj?m_POQ*SUF9O vlWm_7v&KFIYW},8S_ ze|իLovG?pY;>C|ivz+KǺM_P}ZYZDԡ[cmOtҫ_XjU c,- }ş? yW=2K7]~zL6u.a>l|/gtfW/?>k]&W]|>l_@~>#]֙s.׃a=~ ϶ Υ 7I{UQ_~nUǔ U+UJU**ǚ QEVJK-V澪 u/? )֒Z#0r,<_v >wkm_zJᖏKK{A{0,\oUyM-uZMcviVU+ZҵկV/+cRzVMޑVErt_W?»k;_;j -|BhlUBǞkYWAv#ǺQcR֥N]OYP,VZ?e'Nnkҍh%puRGXQEU*Ti+EV@s77U5?kK?/Vfwm_/c>>?Ҝ|J5T{I;|z?zҾ7m>_??wzW>~$׭.|G[xv]tZGt։hxX_|2򖽿2iIy-g kd/?֏aͶOcK F5+bݚ5:F-:WWMxƖMѯ|? 3XƩά-3T +bUY{HuVsgk?ix:{u񦑦jV?gb(:r*֧d{o&tv_dϱWĝsY[k'UΧJ-is_f|qޙZk,$e35 ϶Z׏cαV}k"_Nֻ_xĚ]ޙ[\jw\nJGLgzi^}gU5FewmoV/EH^Jgcț}}kVNz^_b#Vrg>/x_?f_F\n~uK_~겿g ¯ҭֻ[].S妽]<׃}4o]=gq鷖wV/k[;_ӭu?m?d}Kon\W?ZVFW}L\qS-=I&6 wkkKh{j>I_~^//fZǖJGL5Z][bQFF]e>E}O>Үt۟Ul\r1qr_3V| iE}x{>S$uo 7;>˦jֻ']T#hV.iѧk8j:j1{=? ;˯j<gi᫛Z]2}ʼ=Afs˯km<7s!?˖|SަϏn?o􇾂}1-r?NJ--UWởi~W~U~v]ҽSȧ+zM/GY_-m>u[KJ%N\?wՍZIv{+ (-JU(2( O^[^x7J_>$?5kϥq_|Cg 'c?Utj~h~}bjF]55~Aҭխ,u^JCMu=(E[UYVlmh=tF<ۖu{k;oW}]?֏k?5:Q{՞m~G߱^n^ci~W:-ȹMVƹVt[o􋏲E]PjEJG:h]#SmV^}~k:ƛgsk}*7iՙN\M:ٿSj<1ksqW6BeٿO_i~.y(s%$Jsl/_7V}RVuu_?ׅҫ&MnQEy'QE[nPeOQEGğ(̹MOvO%ݭ|_J'mK[KkoH-^Ռh|kֽWIL-y>wy@k>&ClUZ[\Vmj 茮yW [zQ[j~Ӟ&u[iސ-/>JM*߇my5c}V|NTӱKş4 hWWWX;o?ס]bOͥݯk҇[;s{Q宼?oZ-Z2NToxcO^kL[j<;[i6կ)U/ht2,;kǵZ\yPmJ]Z&Wm5өj]?ڭZxֹqǵ=W!a%=x3}Vo-kj֭\]UVfZQ֩X(;QEkUJU+j (5 G\eFMS3Q_/-}ƞ?Qiϵ}k}ֽ:~sVWGmiy[kFqVoVmһϝyiZŵjď:ͥ$/i5Q)O*>ضjK| UOx'2 $CKuu_ |TN>ԭ-WbGW+կ?js!LoO|T~7ww6^7ޥTN?=?ڪğx]?ֹkV+ܶg(QE\ǾTtQP:)~П~gOjUEN >Vbj[jV:/54bݥ{Į*?<ismvᵓYyGK/+ʼ=lZki]<nXz^kDkr*YQ$v9I4˯}.m<7s%jdӵkZGi?Bեj.dɻǺMu~%[k^ii5t!)siyZ֗-uv$yLec~mG--~o^i j.am?5s]:o[}/.|IRtFJnG_Z^)'J%*QErQEnU+Q9.PnK;˯[[\(x^_Ov<L猴]FWiOu_kY<^]-x?&~%K_[ZWw?f%^/wmmmO˥tGo>fj?OUNe|ʱyNjzziig\g]y.?vnzP4.c%s٫nuV(Nix\O뗗_f>mJk)rr]zիM* K?|a)s~;ZɴJ_.{oy܏L\˷jBS tw81ucFX-j6u^k_U|yW &m!]2^־U5͵ͧ}חJ\GE~)+uUJU+>2 (~g@QEWIumLJ+u+u⻛K_8~м ZM[WrO߮O3.,aO φ_嵦+_@WfZ]'W{fsykissmtG+Y7&Tv};mcK%=ן?kOJ.+;[O\]--k+ZV}=-~^>j2ƩwssmmV\]sj3u{kLiٺ<]ݭny?WWu?~ލn>/ ȮZ2ֻ_>.vj?^MyWyW<;]]}S[qxkEՇM[ *v5?utZY׋9OջMu}#PEKK:Vg?~/uf%OIaǧ}]Wx.=k|vښڵ+WY[2⾋ GӳVg{[g 'H>xkľҭ|EkmiBZ)O\׭jZfcN%(6>`ssKnx?\ٵ6H,ͥͮڴnCh~0Юm.{]W_+Z>ϫf/ %B~~[ZYQwg_Zx6w?j.iW :v\[\یtr`*u3\6>Iђ0hwyiif=GWi{EymsoW5o"ץ\a smkqȯONܨxFm&ς>vz[/]\\ }֣7~msu|G_ڗ&84v(m77_d%վr.f|ݏҹzS+ojtܮ[}쾔6|9hPɤ_ϵյt }O֓ksj__eG6"z_/.˟Eꋿ S7%v_%:d}ruGV,o恤NK*U/ukϱ 9%(-ҾOY!L׭-ڼ>/_Z~74rc1O'?_>0w{[/JOtW]n.x_|Cp.m:&\+>f5-;znW|$"Դksosŷ[G/-s.-,n|9iWw>×7zp.{[ 7U*Vm5󢾄ic/qEdKS̚\j676=kOX~Uϋ:mZ>~1qGdٮ>>՘6 fx+7FЅŽ?뺅Λucy4 c&'_4&gsj jrMki~"Hx2:K p/#.|=jl=R8/xvLm68Wp}}*< 7+nu]?Đi-qp.-zPmSdx>ٯ]w>_On+s{m+VÖvw?5GO~l*.ܬ4i~ ӹK6N]JWÛ]SMҮ}5:ͱԴSv78c(>b7B*gG\xqk~r>i_jsj:jJ~-z˩TgO6W|;'¿mumsV:MV+.nkil7ڮnk+On,nnDRkփ#y^r{g@n<mg+[}ڽ+ڕy w6zժb;ǚ{Oϯ}֏7 mQE%:?.nn_kjSmh\lb} ғ^GPE㥭5gJ]7M &ۀ* ŷ66-7m!4yOnthbW8Xԭ]w|c9<ֿ|GYxf6Zvךnmͽ VO ׺m6t s@Jq]sAۚ+?mmE#_kv]5>Ϸ5]xCM:?i??:RRᴫKqpGٺQEM)O__/4[Z4 Nͽkr f^xsRnn.~EAZ͖O>qU(IE!)fgqn7%q\hVVi!iysV=y[]fJ䟳U6Z* k>)vca6t\v힕W_y\N*ׇ!Yh]\m'OٽLWkV:mɶڷ ꮵ}<nAm0k\gLu~3q@8>:ԭ[[[[k[m^}i} [gڧn4sTt˭H?kj.ӣ7w6Wu:Jrli\Ftz5c< jMm.?]*ςAMɹzY-5!)Wtͽ=q^3k+~?uK#'Mk%Vw;MN=onyRhUw BWkjzH|> '>}ݷ4Z[7uz[7ڇ{~ Z4!o`g9-mkmqm;N}koGԮGyIf+m_˜Iݜ֫\[x$]]~ԃXczWHJΟ?CΏ4Pv{_#iY?j5/ MxsimcI]BQ9SռEoki\ܓ[սĖ m*HjVՏPb#9ce& "u@[rjAy}:}䍾gK˝4vgsM0]b> rRkvZi8|'\|Cu} S;ѨGMot3A*F];S,0uZN[LVW:XOg9¾,ӮOe6އY|% ք^~4𥶗^E_LsP7ޥx\k8W@֛www6ԋ0m IO mGrZ..5[v~ HV]TEiq7\UI/\Q(°ݯY1[9Dpݱ>t3nvknnE=·:lLyK5sno}xV(lů 1)uw\`==i-*[1{Mn +"vOgg48ޛ6=gxr>&ci$>}O/cXồ.R%oF{Quj֙];O ƒ\{SIad\"T:犤ѭ]9=Ŕ5ܶiumt:? }?`6}8']O-r_[Q^C?C(`}Dd9J  c &A?2K?@ABCDEFGHIJKLMNOPQRSTUVWXYZ[\]^_`abcdefghijklmnopqrstuvwxyz{|}~      !"#$%&'()*+,-/012345:=^?>@BACEDFGIHJLKNMOPRQSTVUXWZY[\]_`vabcfdegihkjmlnoqpurstwzxy{|~}Root Entry FG><@Data .nWordDocuments[ObjectPool >G>_1075186896F>>Ole CompObjeObjInfo  !"#$%&'()+,/23456789=@ABCDEFJMNOPQRSTUVX[`cdefgknopqrsw| FMicrosoft Equation 3.0 Equation.2 Equation.39q `AJ e FMicrosoft Equation 3.0 Equation.2 Equation.39qEquation Native <_1075186894 F>>Ole CompObj eObjInfo Equation Native  <_1075186893F>>Ole   `AJ e FMicrosoft Equation 3.0 Equation.2 Equation.39q `AJ eCompObj eObjInfo Equation Native <_1075186892F>>Ole CompObjeObjInfoEquation Native < FMicrosoft Equation 3.0 Equation.2 Equation.39q `AJ e FMicrosoft Equation 3.0 Equation.2 Equation.39q_1075186889F>>Ole CompObjeObjInfoEquation Native <_1089791443#F> h>Ole CompObj!f `AJ e FMicrosoft Equation 3.0 DS Equation Equation.39q { s .  @ &@  & MathType`ObjInfoOlePres000 "Equation Native *_1089791463%F h> h>Times New RomanLSwUSw0-2  )y 2 ,...,2 N1.2 (.2 ),2 ,,2 4(,Times New RomanLSwUSw0-2  n,2 hi,2 y,2 x, Times New RomanLSwUSw0-2 ei,2 i,Symbolw@F _LSwUSw0-2 @=, & "System 0-%ip (x i ,y i ),(i=1,...,n) FMicrosoft Equation 3.0 DS Equation Equation.39qOle -CompObj$'.fObjInfo0OlePres000&(1"4  .  &@ & MathType0 Times New RomanLSwUSw0-2 dyTimes New RomanLSwUSw0-2 Ry2 LxySymbolw@ (LSwUSw0-2 6y & "System 0-%"" x"R d FMicrosoft Equation 3.0 DS Equation Equation.39qF4  Equation Native :>_1075186878@+F h> h>Ole ;CompObj*-<fObjInfo>OlePres000,.?Equation Native G1_1089791512F1F h> h>.  & & MathTypePTimes New RomanLSwUSw0-2 `Ry2 `^yySymbolw@ LSwUSw0-2 `Ny & "System 0-X,$ y"ROle HCompObj03IfObjInfoKOlePres00024L FMicrosoft Equation 3.0 DS Equation Equation.39qf4 A .  &` & MathTypePSymbolw@ ELSwUSw0-2 `cdySymbolw@a LSwUSw0-2 `]+y2 `m=yTimes New RomanLSwUSw0-2 `)y2 `I(yTimes New RomanLSwUSw0-2 `xy2 `ry2 `^yy & "System 0-%) E # y=r(x)+Equation Native WE_10897915447F h> h>Ole YCompObj68Zf FMicrosoft Equation 3.0 DS Equation Equation.39q%hl r(x) FMicrosoft Equation 3.0 DS EqObjInfo9\Equation Native ]5_1075186876<F h> h>Ole ^CompObj;>_fObjInfoaOlePres000=?bbEquation Native h)uation Equation.39q:  .  `&  & MathType0Symbolw@F LSwUSw0-2 `"dy & "System 0- o/  FMicrosoft Equation 3.0 DS Equation Equation.39qX4  .   & & MathType0 _1075186875c:BF h> h>Ole iCompObjADjfObjInfolOlePres000CEmEquation Native t6_10897915825KHF h> h>Ole uTimes New RomanLSwUSw0-2 c2ySymbolw@ LSwUSw0-2 sy & "System 0-3  2 FMicrosoft Equation 3.0 DS Equation Equation.39qCompObjGIvfObjInfoJxEquation Native y)_1089791614MF h> h>% pL x FMicrosoft Equation 3.0 DS Equation Equation.39q4   .  &Ole zCompObjLO{fObjInfo}OlePres000NP~B & MathTypePTimes New RomanLSwUSw0-2 `0)y2 `,y2 `H(ySymbolw@? LSwUSw0-2 `wyTimes New RomanLSwUSw0-2 `xy2 `fy & "System 0-Equation Native =_1090306723SF h> h>Ole CompObjRUf%! E% f(x,) FMicrosoft Equation 3.0 DS Equation Equation.39qxR ) .  ` &  & MathTypeObjInfoOlePres000TVzEquation Native _1089791633/YF h> h>Symbolw@/ 5LSwUSw0-2 ,y Symbolw@ LSwUSw0-2 =ySymbolw@/ 6LSwUSw0-2 =y Times New RomanLSwUSw0-2 my2 jy2  jy2 % jyTimes New RomanLSwUSw0-2  xy2  gy2 xy2 fy Times New RomanLSwUSw0-2 >1yTimes New RomanLSwUSw0-2 m )y2  (y2 ))y2 ,y2 H(ySymbolw@ LSwUSw0-2 wy2 wy & "System 0-  f(x,)= j g j (x) j=1m " +b FMicrosoft Equation 3.0 DS Equation Equation.39q  S .  ` &Ole CompObjX[fObjInfoOlePres000Z\@  & MathTypepTimes New RomanLSwUSw0-2 / my2 jy2 xy2 Lgy Times New RomanLSwUSw0-2 `jyTimes New RomanLSwUSw0- 2 ,...,2 w1.2 9),2 (,Symbolw@ LSwUSw0-2 q=, & "System 0-%OH\ g j (x),j=1,...,m FMicrosoft Equation 3.0 DS Equation Equation.39qEquation Native k_1089791678_F h> h>Ole CompObj^afObjInfoOlePres000`bEquation Native Q_1075186869eF h> h>< 4Z - .  `&  & MathTypePTimes New RomanLSwUSw0-2 `))2 `q,)2 `'()2 `,)2 `$()Symbolw@ KLSwUSw0-2 `w)Times New RomanLSwUSw0-2 `x)2 `af)2 `y)2 `FL) & "System 0-%5pT L(y,f(x,)) FMicrosoft Equation 3.0 Equation.2 Equation.39qOle CompObjdfeObjInfogEquation Native < `AJ e FMicrosoft Equation 3.0 DS Equation Equation.39q% y .   "&_1089791690]njF h> h>Ole CompObjilfObjInfoOlePres000kmEquation Native _1089791730pF h> h>Ole !@ & MathTypeSymbolw@ LSwUSw0-2 "] y2 ] y2 ] y2 j-y2  -y2 y2 -y2 ! =yTimes New RomanLSwUSw0- 2 = otherwise2 xy2 Afy2  yy2 xy2 #fy2 yy2 hif2 kxf2  ff2 zyf2 FLfSymbolw@ LSwUSw0-2 5ef2 wf2 !ef2 wf2 wf Symbolw@' ,LSwUSw0-2 efTimes New RomanLSwUSw0-2 |f2 )f2 Q,f2 (f2  |f2 |f2 )f2 3,f2 (f2 |f2 0f2 ))2 ,)2 ()2 .,)2 () & "System 0-%eDV L  (y,f(x,))=0if|y"f(x,)|d"|y"f(x,)|"otherwise{ FMicrosoft Equation 3.0 DS Equation Equation.39q\8  .  @`&  & MathType-  Symbolw@CompObjorfObjInfoOlePres000qs`Equation Native  !$%&'()*+-03456789:;<>ADEFGHIJKMPUXYZ[\]^_`abcdefghijlmpstuvwxyz{|}~q ^LSwUSw0-2 y Symbolw@ LSwUSw0-2 g=ySymbolw@q _LSwUSw0-2 h=y Times New RomanLSwUSw0-2 ony2 iy2 iy2  iy 2 4empeTimes New RomanLSwUSw0-2 \xm2  fm2  ym2  Lm2 nm2 FRm Times New RomanLSwUSw0-2 1mTimes New RomanLSwUSw0-2 *))2 ,)2 ()2  ,)2 > ()2 1)2 ))2 ()Symbolw@ LSwUSw0-2 w)2 Sw) Symbolw@ LSwUSw0-2  e) & "System 0-%»j, R emp ()=1nL  (y i ,f(x i ,)) i=1n " FMicrosoft Equation 3.0 Equat_1075186865)vF h>>Ole CompObjuweObjInfox ion.2 Equation.39q `AJ e FMicrosoft Equation 3.0 Equation.2 Equation.39q `AJ eEquation Native  <_1075186864{F>>Ole  CompObjz| eObjInfo}Equation Native <_1075186863yF>>Ole  FMicrosoft Equation 3.0 Equation.2 Equation.39q `AJ e FMicrosoft Equation 3.0 Equation.2 Equation.39qCompObjeObjInfoEquation Native <_1075186862F>>&Ole CompObjeObjInfoEquation Native < `AJ e FMicrosoft Equation 3.0 Equation.2 Equation.39q `AJ e_1075186861~F>>Ole CompObjeObjInfoEquation Native <_1075186860F>>Ole CompObj f FMicrosoft Equation 3.0 DS Equation Equation.39qF{  .  @& & MathTypeP Times New RomanLSwUSw0-2 2yTimes New RomanLSwUSw0-ObjInfo"OlePres000#"Equation Native ,F_1075186859F>>2 r||2 "||Symbolw@ LSwUSw0-2 w| & "System 0-*(Խ |||| 2 FMicrosoft Equation 3.0 DS EqOle .CompObj/fObjInfo1OlePres0002uation Equation.39qb 1 .  `&@ & MathType` Times New RomanLSwUSw0-2 *yTimes New RomanLSwUSw0-2 o,y Times New RomanLSwUSw0-2 iy2 iySymbolw@ LSwUSw0-2 xy2 xy & "System 0-5(  i , i*Equation Native =Q_1075186858F>>Ole ?CompObj@f FMicrosoft Equation 3.0 DS Equation Equation.39q4   .  @& & MathTypePTimes New RomanLSwUSw0-2 `-ny2 `.iyTimes New RomanLSwUSwObjInfoBOlePres000C4Equation Native LE_1075186857tF>>0- 2 `,...2 `1.Symbolw@ LSwUSw0-2 ` =. & "System 0-)hoD i=1,...n FMicrosoft Equation 3.0 EquatOle NCompObjOeObjInfoQEquation Native R<ion.2 Equation.39q `AJ e FMicrosoft Equation 3.0 DS Equation Equation.39q a _1089791800hF>>Ole SCompObjTfObjInfoVOlePres000WEquation Native k_1089791805F>>Ole n.  @`&  & MathType- @ (  Symbolw@$ bLSwUSw0-2 !y Symbolw@m LSwUSw0-2 =ySymbolw@$ cLSwUSw0-2  +y2 +y Times New RomanLSwUSw0-2 ny2 \iy2  iy2 2 iyTimes New RomanLSwUSw0-2 ny2 Cy Times New RomanLSwUSw0-2  1y2 *y2 2yTimes New RomanLSwUSw0-2 )y2  (y2 ||2 ||2 Z2|2 T1|Symbolw@m LSwUSw0-2  x|2 X x|2 ^w| & "System 0-%šFL 12|||| 2 +C( i + i* ) i=1n " FMicrosoft Equation 3.0 DS EqCompObjofObjInfoqOlePres000rEquation Native Wuation Equation.39q  .  @`&  & MathTypeSymbolw@ yLSwUSw0-2 :y2 :y2 :y2 B:y2 :y2 @=y2 @y2 n +y2 ! y2 B-y2 t +y2 ' y2 -yTimes New RomanLSwUSw0-2 @ ny2 @iy2 yy2 xy2 Bfy2 vxy2 fy2 $yy Times New RomanLSwUSw0-2 xiy2 aiy2 `S iy2 `Ciy2 `Xiy2 Y iy2 *iy2 iyTimes New RomanLSwUSw0- 2 @ ,...,2 @ 1.2 @,.2 @0.2 @*,.2 r).2 ,.2 (.2 D).2 ,.2 (. Times New RomanLSwUSw0-2 *.2 ~ *.Symbolw@M LSwUSw0-2 @x.2 @x.2 z x.2 K e.2 Fw.2  x.2 Q e.2 w. & "System 0-%;{dI y i "f(x i ,)d"+ i* f(x i ,)"y i d"+ i  i , i* e"0,i=1,...,n{_1089791825F>>Ole CompObjfObjInfo FMicrosoft Equation 3.0 DS Equation Equation.39q(   .  ``&   & MathTypeTimes New RomanLSwUSw0-2 )yOlePres0002Equation Native _1076143861F>>Ole 2 L,y2  (y2  )y2 (y2 )y2 H(y Times New RomanLSwUSw0-2 " *y2 1yTimes New RomanLSwUSw0-2 xy2 xy2 W Ky2 xy2 fy Times New RomanLSwUSw0-2 iy2  iy2 iy2 ny2 iy`Times New RomanLSwUSw0-2 7QSVSymbolw@ ILSwUSw0-2  aV2 aVSymbolw@ LSwUSw0-2 -V2 =VSymbolw@ JLSwUSw0-2 V Symbolw@ LSwUSw0-2 G=V & "System 0-%¬i f(x)=( i=1n SV "  i " i* )K(x i ,x) FMicrosoft Equation 3.0 DS Equation Equation.39qCompObjfObjInfoOlePres000fEquation Native LF>  .  @& & MathType-Times New RomanLSwUSw0-2 ny2 lCy Times New RomanLSwUSw0-2 yiySymbolw@ jLSwUSw0-2 `y2 `Jy Times New RomanLSwUSw0-2 *yTimes New RomanLSwUSw0-2 `40ySymbolw@ LSwUSw0-2 `Yay & "System 0-0X  0d" i* d"C FMicrosoft Equation 3.0 DS Equation Equation.39qF s .  @& & MathType_1076143870F>`u>Ole CompObjfObjInfoOlePres000Equation Native H_1074667375F`u>`u>Ole -Times New RomanLSwUSw0-2 ny2 lCy Times New RomanLSwUSw0-2 qiySymbolw@ LSwUSw0-2 `y2 `FySymbolw@q LSwUSw0-2 `RayTimes New RomanLSwUSw0-2 `40y & "System 0-," 0d" i d"C FMicrosoft Equation 3.0 DS Equation Equation.39qCompObjfObjInfoEquation Native :_1089791843F`u>`u>4 n SV FMicrosoft Equation 3.0 DS Equation Equation.39q%Ÿ_# K(x,x i )=g j (x) j=1m " g j (x i )Ole CompObjfObjInfoEquation Native _1059140843F1>1>Ole ObjInfoEquation Native < `AJ e FMicrosoft Equation 3.0 DS Equation Equation.39q  _1074667570F`u>`u>Ole CompObjfObjInfoEquation Native )_1074667580 F`u>`u>Ole CompObjf FMicrosoft Equation 3.0 DS Equation Equation.39q(ItI  FMicrosoft Equation 3.0 DS Equation Equation.39qObjInfoEquation Native ,_1089792025W>F`u>`u>Ole hookxEO;n1}3泫Q$(MB`/@ VP@ mGPq #Ieϛ{ Phy"&ZO/Vj.<ۼ,[^_+r"zFzWFdb^~(Dd ,h   s *A? ?3"`? 2l;2H#H`!@;2H#Bxcdd``Vdd``baV d,FYzP1n: ,B@?b X ㆪaM,,He`PI? 01d++&1-bF: @6°$R 1y1ps0Nb0D@(\PՆ˝s暁]v=pĽV \D.v$+KRsA<.f8` Dd l@N  s *A? ? 2Y_M)8vGjP5`!-_M)8vGjPp xcdd`` @c112BYL%bpu#d؅* b;FLLJ% H  b#3DAODd @`h  s *A? ?3"`?2sF ^Rg&,o`!gsF ^Rg&, 5xcdd``Ned``baV d,FYzP1n:&B@?b u  ㆪaM,,He` @201W&0;EUy}1#X r8WcH& @c112BYL%bpu5Zs'º:ZH7a]#EKA-8U"S'_zb:4:zmn0? nK$C4268%vb%:pQ <dfoAR,h .I2Dd  @h  s *A ? ?3"`?2v(fe GjR'`!J(fe Gjb8 xcdd``6dd``baV d,FYzP1n:&B@?b u 00 UXRY7S?&,e`abM-VK-WMc Z `  @Ct mĕѬ'1:p?d_penR~CPt1H23TT@e{ l%\L61TUrAc c]ؑĤ\Y\ t0.?0f}aDd  `h  s *A ? ?3"`?2.b tdHz*`!.b tdHz@`Pxcdd``eb``baV d,FYzP1n:~! KA?H130@0&dT20| 01d++&1-bF: @6j`r$U9At8Oا<VTBygAqU9!|t+ssa\ q# p2P{gCF7&O1-s{lE1ylRa-IJgR o9n_dADQ"#%1`Ɨ5r= |.h.S v 0#I)$5b2B16?2նBaDd h|Lh  s *A ? ?3"`?2^s÷kEaP],`!y^s÷kEaP]@ `0Gxcdd``.gd``baV d,FYzP1n:&.! KA?H1Zq10sC0&dT20 `[YB2sSRs|1#X r8WLn%g> ~z%>?} $b.#(} ]===@ `|fP4hn& Ma(w D0qM&Ώ ά ha"׏0s@**t) z v 0#Ĥ\Y\ tevIDd (@Bh  s *A ? ?3"`?2SZh;g6if)^ѡLF~ـ.>q`q(xsMNoLߢ;Eï=l3|m0V$V]j]5{& ׏y LNU:`+juN.FߛI۞`ԲV;J8$?óCY LZgϷ Wl=Qtt3!?t y<7~ /#f+'XU;^x >%R~8QzOj5%YûgxPu? ]UiM>y yO*-e cs D}SGΓ Dd  Lh  s *A? ?3"`?2Qr_㶈~]ѯl-6`!%r_㶈~]ѯl@Ptxcdd``>$d@9`,&FF(`TɁAXRcgbR 6GfĒʂT @_L ĺE@,a +LBTN9 +ss3v-bF:`n2-c<y۸|?}lRa 5wNLPs|XE$47$37X/\!(?718b7'pɄ0G@P>#0ZRs R}L0}<"-_ W y&'py7?(K SbC0K SS  t| U3|?4pxXQ̿Q%ޘX0s?q4r=n>a0G#.09=LoKp8WhrA*.pyK.F&&\ Z~1S.?0Dd9J  c &A?2KC`!@I-ً0zԜ@p|xcdd`` @c112BYL%bpu|÷y)sD 3X~߇*Zq|k=emS˽wJܓ^M59>ǏTk'ZgSI ΅GiB+x wO_js 'yYghs-}<ޓty\ŁSoޢr7+FB.ͭ#HWbũQ`!h 2qI>WbũR `hPxcdd``^ @c112BYL%bpu_Ch .0 d4+a|)~_w5&nTYl3P=!*'Ͻ`_sThAcY V&Rb#<|ãK>)R>J"*;2 ,ytHaxHTx#'JdaE]p3 Ma(wzOrT$ |,͞3\\%Ĺ1 ]= -Ԟi8a__JXX!傖>\ b;RcdbR ,.I@X`sTDd `|$h ' s *A? ?3"`?&2iΧ}@] tT`!liΧ}@] `X 0:xcdd``ndd``baV d,FYzP1n:&v! KA?H1ā깡jx|K2B* Rvfjv,L ! ~ Ay >_C΀H9@ڈk,Vs%<?e`eȬDw'(P 27)?(𒛫;!2[Rd@04ݯ$3M@* * Z2sL \NQ]=4Ĥ\Y\ t0.?0̱~IDd `|8h ( s *A? ?3"`?'2Qi]|RiV`!aQi]|R`X 0/xcdd``ed``baV d,FYzP1n:&v! KA?H1ā깡jx|K2B* Rvfjv,L ! ~ Ay >_C΀H9@ڈk"Vs%<?e`fȬDw'(P 27)?(+ ;!2[Rd =SL OR8߄  \%F&&\, doDd |hf N ) s *A? ?(2B*/.vҴ7_Hi2Y`!*/.vҴ7_Hi͒`@0|xcdd``> @c112BYL%bpu% #RpeqIj.n P>Dd Ih * s *A? ?3"`?)2٤S'Hh5Ďi[`!٤S'Hh5Ďiע`Pxcdd``db``baV d,FYzP1n:B@?b 30@0&dT20| @201d++&1-b @6j`RhV^ ΍*U~* &_/bW?ye~ϼ\Fb.#L@*2+a< ~Alon%oِ0UqeX #Eo`: tD8s@|o4{01P> LWnO#*ߍ‡_gBܤrD>&Lp?Ŗ0̍`*'k[%asb+b.47ssbl@/>um* JVq_ԝ1AR!e=rm ։Rnu3]uwbnE0_Yf\}O!=-JDd9J 8 c &A?72Kg8=`ؾ'jv`!g8=`ؾjxcdd`` @c112BYL%bpubl@/>um* JVq_ԝ1AR!e=rm ։Rnu3]uwbnE0_Yf\}O!=-JDd9J ; c &A?:2Kg8=`ؾ'j2|`!g8=`ؾjxcdd`` @c112BYL%bpu`u>%¼@ |f(x)|d"|( i=1n SV "  i " i* )K(x i ,x)| FMicrosoft Equation 3.0 DS Equation Equation.39qOle CompObjfObjInfoEquation Native %°# d"|( i=1n SV "  i " i* )|"|K(x i ,x)| FMicrosoft Equation 3.0 DS Equation Equation.39q_1089792168F`u>>Ole CompObjfObjInfo   !$'(+05:?DILORUX[`cdgjmruvwx{~%vP d"C"| i=1n SV " K(x i ,x)| FMicrosoft Equation 3.0 DS Equation Equation.39qEquation Native _1089792201F>>Ole CompObjfObjInfoEquation Native _1089792263F>>Ole  %¢@ K(x i ,x)=exp("||x"x i || 2 2p 2 ) FMicrosoft Equation 3.0 DS Equation Equation.39q%7l K(x i ,CompObj fObjInfoEquation Native S_1089792283F>>x)d"1 FMicrosoft Equation 3.0 DS Equation Equation.39q%I f(x)d"C"n SVOle CompObjfObjInfoEquation Native e_1076144069F>>Ole CompObjfObjInfo FMicrosoft Equation 3.0 DS Equation Equation.39q0K, n sv FMicrosoft Equation 3.0 DS Equation Equation.39qEquation Native :_10897923219F>>Ole CompObjfObjInfoEquation Native  L_1090218736F>>Ole "%0G Ce"f(x) FMicrosoft Equation 3.0 DS Equation Equation.39q }>\ C=max(|2y+3 y |,|2y"3CompObj#fObjInfo%Equation Native &_1090218909Q F>> y |) FMicrosoft Equation 3.0 DS Equation Equation.39q h 2y FMicrosoft Equation 3.0 DS EqOle )CompObj *fObjInfo ,Equation Native --     ! "$#%'()*+,-./0123456789:;<=>?@ABCDEFGHIJKLMNOPQRSTUVWXYZ[\]^_`abcdeghijklmnopqrstuvwxyz{|}~_1074667937F>>Ole .CompObj /fObjInfo1uation Equation.39qHI  y FMicrosoft Equation 3.0 DS Equation Equation.39q h 2yEquation Native 26_1090218940F>>Ole 3CompObj4fObjInfo6Equation Native 7-_1074668152/F>>Ole 8 FMicrosoft Equation 3.0 DS Equation Equation.39qp 3 y FMicrosoft Equation 3.0 DS Equation Equation.39qCompObj9fObjInfo;Equation Native <:_1074668185F>>Ole =CompObj>fObjInfo@Equation Native A1@#< " FMicrosoft Equation 2.0 DS Equation Equation.29q // s_1062827304"F`>`>Ole BCompObj!#CfObjInfo$EEquation Native F<_1074668251*'F > >Ole GCompObj&(Hf FMicrosoft Equation 3.0 DS Equation Equation.39qKX3  y/x2 " 2 n FMicrosoft Equation 3.0 DS Equation Equation.39qObjInfo)JEquation Native Kg_1074668264,F > >Ole MCompObj+-NfObjInfo.PEquation Native QG_1074668296%1F > >+# " n  FMicrosoft Equation 3.0 DS Equation Equation.39q;hug = lOle SCompObj02TfObjInfo3VEquation Native WWnnn  FMicrosoft Equation 3.0 DS Equation Equation.39q%  =3 FMicrosoft Equation 3.0 DS Eq_1089976431k6F > >Ole YCompObj57ZfObjInfo8\Equation Native ]1_1089792608;F > >Ole ^CompObj:<_fuation Equation.39q%e@ (x i ,y i ),(i=1,...n) FMicrosoft Equation 3.0 DS Equation Equation.39qObjInfo=aEquation Native b_10897926404@F > >Ole eCompObj?AffObjInfoBhEquation Native iE_1089792661EF > >%)wD& y=r(x)+ FMicrosoft Equation 3.0 DS Equation Equation.39q%PKw r(x)Ole kCompObjDFlfObjInfoGnEquation Native o5_1089792768CfJF > >Ole pCompObjIKqfObjInfoLs FMicrosoft Equation 3.0 DS Equation Equation.39q% K R pred ()=1n test (y target "f(x test ,)) 2i=1Equation Native t(_1075098181OF > >Ole yCompObjNPzfn test " FMicrosoft Equation 3.0 DS Equation Equation.39q{ r(x)=asin(x)xx"["10,10]ObjInfoQ|Equation Native }_1074668740TF > >Ole  FMicrosoft Equation 3.0 DS Equation Equation.39q   a FMicrosoft Equation 3.0 DS Equation Equation.39qCompObjSUfObjInfoVEquation Native )_1089976678pYF > >Ole CompObjXZfObjInfo[Equation Native i%M( 1,10,0.1,"10,"0.1 FMicrosoft Equation 3.0 DS Equation Equation.39q  a_1090159797W^F@ >@ >Ole CompObj]_fObjInfo`Equation Native )_1090159909cF@ >@ >Ole CompObjbdf FMicrosoft Equation 3.0 DS Equation Equation.39q   FMicrosoft Equation 3.0 DS Equation Equation.39qObjInfoeEquation Native )_1089794264hF > >Ole J = c &A?<2Kg8=`ؾ'j`!g8=`ؾjxcdd`` @c112BYL%bpu c &A?=2Kg8=`ؾ'j`!g8=`ؾjxcdd`` @c112BYL%bpu2Kg8=`ؾ'jփ`!g8=`ؾjxcdd`` @c112BYL%bpuXVwRF9pϵ'LT`*_U\MO "&_ktpLb{V [7Zxqo OJ0<&_pTw{GA__ ixc e'3\/*thVI}`!]>hVI}b`XPxڭK@߽I N"EPT chk [+5PPpItv"S:hxwIV?R t= zFmr"/+rEQ`d)WhtNG"5nN 3mq:׳N|,ē[LRυE:x- zW8'ǻA}'c"zeɿCr?qw~WѸKx7G] ?1?骝e[\W %7,c֭O~HW2Zs?=eeW60C'P-oZqB0>uu+oJ: .-y+?'h;mx(PedD_%Uk/S!")k'ofU2xW$>༴Dd Dh C s *A? ?3"`?B2-|Q"A@y].;/Wu0-GgGKdk1`j'1ee3y~ڡSDd h&h E s *A ? ?3"`?D2gZw.H]s`!kgZw.H]`@hn |9xcdd``ed``baV d,FYzP1n:&v! KA?H1Z πqC0&dT20$ KXB2sSRs|1X r8Wc HưƟk _F%P 27)?(p v$wT,F3w7*pa!(${2HE&F0R~7 J.h*.] HaĤ\Y\ te,xmDd <h F s *A!? ?3"`?E22wI`!2w  hSxcdd``e 2 ĜL0##0KQ* W)d3H1)fYˀP7$# ( XXbS%̤Kf%3I?DYB2sSRs|1#XFF\ LA^ < T 3Oð@(\PbLp{^s_w BnO 84@b lOE4qSD.v$+KRsA<.h*.?1YzDd9J G c &A?F2Kg8=`ؾ'j`!g8=`ؾjxcdd`` @c112BYL%bpu @c112BYL%bpubq`!i+pțJG>b` h7xcdd``d 2 ĜL0##0KQ* W&d3H1)fYˀX (T UXRYP,@1[).%̤a~~DFL ! ~ Ay >_Cj#~# J_רD5i3H[QnF 1 n3HEz s%aBܤr;;vJ.hlrc b;RX121)Wx\f=hEDd |Ch K s *A$? ?3"`?J26C\$ \֟`!6C\$ \`@0xcdd``>$d@9`,&FF(`T)i A?d-$P=7T obIFHeA*CPD.#l) @,a ˙&v& rof01d++&1-b=`gT ?4yQAP/0ܘXQqIr#w] 0~-$&x\Aq *_wf/ܤriVzRQ ǙL hqp% p{p @c112BYL%bpu% #RpeqIj.n .M>HDd ,h N s *A%? ?3"`?M2D 9۫=0wK x`! 9۫=0wKf`!xcdd``$d@9`,&FF(`Tc A?d=bAǀ7$# !( XXbS C*% d2D YB2sSRs|1#XLF]F\q J.!I)$5ズI&0h3X )6YDd |:N O s *A'? ?N2a`*^4J=x`!5`*^4J`@C0xcdd`` @c112BYL%bpuh0qIc7L@'.[+ b;FLLJ% H  b#3X,:2Dd B S S A)? R2,ֵQ6/]'y=`!ֵQ6/]'y4@Yxcdd`` @c112BYL%bpu .| z- `+KRsA<.,BF0( J-fDd9J T c &A?S2Kg8=`ؾ'j`!g8=`ؾjxcdd`` @c112BYL%bpu0_Kx1 Ma(wְr+X@*,`a%4{L6 [+) v 0y{鐑I)$5ウ"#p,yDd!X V c &A?"`U2Kg8=`ؾ'jS`!g8=`ؾjxcdd`` @c112BYL%bpu .| z- `+KRsA<.,BF0( J-f2Dd P4h ` s *A1? ?3"`?_2vdAq~meR`!JdAq~me%xcdd``$d@9`,&FF(`TABRcgbR P PT obIFHeA*/&*de-b Y wf:rȁL@(\Pdz< f22$b|bn7Ś jn?|˝s.`:R[ cE'~( Оbr<Bkh)I-.ALF#K+0?3b3s0fD>PKt@tB \2 \BtA``GI)$5ゖ{ tO`q]Dd lS\ a s *A2? ?"``2vc=,`!vc=,xxcdd``na 2 ĜL0##0KQ* W A?d fĒʂT/&`b]F"L La 0 roB2sSRs|1XLF]F\ LB J;0~H^·* 8@|MN(?dcqfLXE&#O+ssK9c3Q> ]=< 98_ķŰĐ7jO6d~!vd~S 0[ç( f =Ĥ\Y\ qA(0p3XRnDd eN b s *A3? ?a2-$N|j a:Q r`!$N|j a:Q:@`!xEN=Q;w$.*E#T(+.8$$rJ' ?I8/3f,+F 猉2F5^FT)DjD߅uM>: *IxX͓]5ŧ o?lq! aQ:pAC0qW7Jn;1db̔<>iip.VTDd |@jh c s *A4? ?3"`?b2>\L$j{FtA`!l>\L$j{Fb`  :xm1KP&-D8ؐ *bZAqOAAI0޻$#}p{w `@؄^ **5Ey6q 5ֶWV\c~am˘q?cM7?<袰 ^,Q'dx+=ey2sў^Vp`xw ?7za׶ Y 7<?P_0(KFsl'jy?(:vn9 W&Vd)j{h eƊh+ʦ8Ѥo٨Y,sRADd aB d S A)? c2,ֵQ6/]'y`!ֵQ6/]'y4@Yxcdd`` @c112BYL%bpu .| z- `+KRsA<.,BF0( J-fDd9J e c &A?d2Kg8=`ؾ'jW`!g8=`ؾjxcdd`` @c112BYL%bpu: *IxX͓]5ŧ o?lq! aQ:pAC0qW7Jn;1db̔<>iip.VDdX i c &A?"`h2Kg8=`ؾ'j`!g8=`ؾjxcdd`` @c112BYL%bpuRtlga+dRfxa;cQot.E `=R>rA=Ie|ܟ 5eݝa'07ET-DdJ k c &A?j2Kg8=`ؾ'j`!g8=`ؾjxcdd`` @c112BYL%bpuu\ 2fAGc1&ө1gn.d\1#e]rӟ_djfgff[(7Qn4d0.n~@JĦ?pBx+d\80[GߴFɒ#՞\ϖd'˂UCgFOTY|ܹp./Ⱦp?{ Y混کML࣒Ml}i ht"bQ":?VbWXv qw5zVgQ|~JkLP[ͯɯnhO!L'D_LfQJ&*7u$a1ee/0 u2QR(),J%RLDK9L9A9E9CFΐ~&}/H/&=ҳHO%3I7%nH7 tC!ݐnH7 tCm-k^b$ڸmX?Jg7_q20?Vlvi]G6ٱ ~ Kp7 MU ˲dSe3ͷ[$3vnY/-uԋkUWn^[<۞<LRLGqc>lg#q^롮. uj׏_&D{cM:4>hZ*x>yRI {r_/mm~_k45 e=|0_?&O(߃J{0wѭw;X~^Mh5_L}V17G`?=_ٮO1緟W$ӊr?&:IUEF&k~GMI7M^{mk#{q~].n]+~gJjz^Whl]C9q~yQWKPoR&)MW=+ќ/um篟W99ZiPU^׉k{ zݠϘļnS *>>hG#ם?5{z^ץJK.Rץq^F^TK]-Tzuf^g*ΌZ'3 r4\?6Լ?5Bo8[Do -mˏEםߪ'{X_]$8d5.Q5-߲8Bq7>t<*~QJM .kFSeBc%Im5|M5'I iFz׵մ-λmxI9m&k4{G[)4tC.hzn7l~&k4)4{M>h *4&k4B)5Xn M6_H5>LUh{M>h&*4!M>h M%5X) BqH5M5X.7)4!M>hz} M\c&ۭdc?'e.kFSG+4&k4Hqc|4diF=><4{M>h+d4XiV<5Xiu Wh!M>h*9jz{M>C=@y{(5m|M[cB͟c t^5^ M6_C| 6|6Ml~&k4 M|MB~&DH5~M>u^5>&B?c&{]B.kFSRKM6C|t'u Ph{M>h@ShB|<=U[MI.kFSKXdB|Ԟ:*4uw^5{^Qh4X)5VP5X)-VOi+u+4Mv^5S3kԎۋר]4:Whw^5l&k4k+}Ѵ"]7k>hڬ~g;#ޛR igqM!n./_iW(uNq&G&2L3^f2ˌM<ÈϚf,6ORڅ&Us{N+9jYczɋl0e&(/2e7{)ck-,?d-;JǨ<β~h# 33E~Sf̕3P>4K7P>ߙoJ׬ߛubdԑ2Or vk,˲ZR?B<'K<(sa!$IK[yB-tN|UKi"9>2ir Aiz#!zn=XN ^4iR@+2̖\Z2̣XţiHfeTj_2~o%o;;CA0rZƼ#r]os 'G?=ѮiL: 66zW^r~wW˯韏O;O?^b Ru0X/erNW(]5 7U@P"2(? H`<̗v\`t K`t&H` FJ CRa2(xReX0@ɒQAoAwtKczYA/y.#A?) Kq*%ԷzCe}.Y֖ KʃѲ#+RL}Ty5ȑ=PuoWzwS`Ma .<G0b2`93`93`9pXay8,gr)GFU&,gHF(X c`y,OI<7 LX say,/`.奰\R˫ay,&X ay',ݰ\{a=q{ay,W.FNX[` r ,bXy<>3`9 rgXn -a>X ,(YX ,7 -? ˇiAXKw⭴ťa `"Z]LKhjZVo;hu,o[|߅S[X>mi$9!߄G`[@XNr{rEi剰< ay,ur9,;ay,˻a+`+`y7,˻`eX /r9,oͰ`EX^ ˫`_\way>,? s`5g0gr,OI<hX΂X ˃ay,r ,>rX,%YN徰{rX` ,wN;rXna=,gt;rgX`,e<0Fc#$,aTN)s&,? 2JbX."X.X^kaEX^`y#,od$o lK`qɰ<3 7 a,`XM˟>-9 '`GZs88Wr ,/U\Kay ,/%\˅K`y5,Ͱ+`5X~߂.,`|r=X rKXn ˝`7,Va9ű\~TX#ay,r,? ay),/X^ ` 7X `aay,oͰ`y,啰,/BX^ ˋay,σ幰<gS<'X ˣay$,gpXNa<4Xa XN<S`9S`9ű)`y,rjB,1}L~ >X|GZ-e~fzhq߁i:tǚcGۖ2ŵAaMeJM[?15FOowkfZYƱoMK=iߏ%7sQNWo}c7+b$1Όa)̤GFwԥ66O 5܈e>qN%yKaZ=eOB}bٵ7H+Ӹǫ[;i~·?1[1:|_T Mp4H`ޯ3k| m4\-&Hl?ؾ_5q"t1{|tQnlt0{|zP>8Y)~K;{J҉Uc'*tCvv^lW.q1Džl2׮e۵xYdZf߳cWljDdJ v c &A?u2Kg8=`ؾ'j `!g8=`ؾjxcdd`` @c112BYL%bpuiDp MUX|/CˮCtRD9N/Q1֜IE{eѽi)׫ APT+Éu9UqlÎ{ëGeB&e3PjWХE\#[.[OG.Ԭ|eJtد}ZK15ֹq 7|T,펛yfD".Vt5HHUt4ṡqWͱ7+h I͋?Sizz-G9?AODh`ca+ ?O)`o&WN2}i mr~ 5AmQ$teb7 eA9Dh*T UAzh3AG9d(q'~mHoBfGz-ҫ^Hg!}*NH' tB:!NH' tBjQ]&]cU~9D Lb\f>+brDev}W<;6xSf9}k:7N#::dvBٺt bdݮ9vNݔeɼ :[=0n2nuz`q/Ɛzx|%ʐ0V6cK=p3SYIrBkxC.@Ph04 .xnnn}38cD\M@߇ʂüA5Џ^ Ay&=<ߛ.tVAB>ߛݫ],yUֳ=VhtݳyƩmzY!WBTI> <v/T=s&KT]*{=U =2$.v2d+NWt}MgVL~q\5Whbr,ku{TEt+֛e\QdQ崐~@G=8 S1g)cPkL@=o5/Ck=dֵe9uYaھsoq']Ǥe}Uh%$ڢhYjw ]WYV;u{Fkf'Gu7sIñ'C!wwY1Ģg]QƷ$Ϻ"~UO(;OȿgcRSiOS{!7AS\ݯn`\sE}l.}vIP"bP1DdJ z c &A?y2Kg8=`ؾ'j`!g8=`ؾjxcdd`` @c112BYL%bpuk'e=JPFwV߿S]յV~[smΒ;{Tosdjʃ۪U+UZZTUNT.]Tyyڵ޽P={T{V}UW^y8p׮==Og5gfu]a[ƺ3>ٖ;_Ue}Iv|_~pigV=}Y?(z| kjǫ?=_1wf̭.aAߘ["t>D{:3#ͤP7 s]y~< .s):/_%De]A|<ϛa5Uyjt&s|Ol6X=OkL7|:opCGk~m"ep'2կkiLTtSu7v[ws&٧lE̋tOTuN^Mi{Xws&1!O=c6k:8 B~\r6PRIˆQ:TCui:Ӧy67ǬP?!tܡó'L^ O?IP~2vG?{'d~QQaG?O6A'OSqDe~B'аJ~iW`' O?IjPj~7m[L ٧cjsg?wcG?3m)O>$cTCc!/Y@?_T~ls@?`+?<wΐ\?˷S].9_ 7bA} ،hS9.? ]mdѹ=ˏ9Ү ^cmB$]9sYk.[~5P[3dK>ZƖ(wcnhKe&7|7![B׍B$]]C,[5-h-}c&3&dK>w![55eKe&7.+ι%d[1:v fֵ:hc[6O;˞ͱ3/ }l;"}FZԺo8c6ǟ-ڒ[~Z-2㵽Mb X]Ͳ+^k]ѵƮ>7+dW^9V ٕtݯ9uzϯ5ŮW4bנlc|}8lRo1bPhVPoq`S>#s鞖M!1MaU;}XyBI݇ᵓП114h cXwssƾ6ݦ{pg|k`IgےM!p$Qu<Mde-eB%MQ^;i yh6 yaO)d9H>la`Lgm )go(1o_ͭ<_c|9.*k ioc|}8l>Qv/^;i ]Ϝ{5h νXwssƾ6ٵŞA¯-`F*یM!p$U/|3sopgmr9M]}&ȹއ̹G\c܃u snu_i0`S>]c6~Յv^mUwޯ)dБD*jv!WsK2粌iM!= %<4^9Qcy]{[K8L3XM!pv|7nls Zi?9w &d| ΝXwssøq9w \ȹ;qZ3XM!pv٤x/|?wJWuWuWvq9> i-CÒLjI=63{1 l Yy %R^;/?Zs\`yҥ'0gbK*aS94ʹ5WZÜ[k4b[b|;Pb}na:0/gטg9fb$=D85s/"=, Z0Ns< 65fo=҅:zƞ[3]cI1/s8Ni`|}8lR$7E<|J}TȦ}8CG$.k,`L{n .R^2 ';=R'+N򎞔8 mRȦ}8CG$QwxHכĜ])d@"vRgHxI20ʐM!p$\0i-eB?$a\=s~Nc-Xss>6;{EW["_81FW_#.h܃9`6l هk3|㞥VM盺a gHb3~}P= %eB'KÎr<;x!B IM2%J _Ipv٦ y_A{1u=Fc/&u]?J=_v=:qO/gטs=qݥi~o+j| k:޿ ^in.^s5dks ㎯Sw=}'Hbqڴif3g0&F6Tǧkn_>hǗuaS>:rT}A9œ 9u;q|j ~_7#VwI}%dS>#yGy$s܅9F6Ǧ@G\Gh\?F>&z+U'8>DUn;aӜGM!pv|󐏻94|h Yw>+O/G>:y/gטg!~}q_b/4|h /|O'<8>_"QyxTg6%8.gyyVcw cLjo+ķwY]] o/Z()g3ӧgmr9M]}&ȹc<þV1o q:LjaunI\1])z8<y^;1oB6:Җk̹,c%8lSHq|z j 9oӿ$8oc:xDj;̙U`|l هk3_?oa|܋}-y .%[ݿU`|l هk3_??>sz-I]K%vj%:yI/gטg^2*_|zSUԷM!B IWEI`v?$s.˘h .R`؍_}y.ɜ{U÷༁u77gs)d.1^S9\QΏkyŜ͹L)Hqܜ\}b{Ms|۽Ku_)dή1ϼd}|:=8*g>#*狙` +M!oljHZ_{\=|zZn V^rm Kkk={vWo;T﹊sGKp^UHWqw[U- ?Jp9ʯZ~®P] Io%yEד?v+Z@y\!^ryǝȜ>嬻Xϧ?/v?c:g/gטg9*_}Vjސ|mCȦ}8CGUQ2wpK2粌 olSHq|z Ay9?_$8?Ⱥ|=mVӜM!pv٤柗r)+TH}v":`vz@W>_E} ;}8CGUQq/ɜ2^!TM!oljH%gI+/yK}^V)dБļdT~deSD6T'v#\1'"=yIM!p$%kkk#9o`ۙ3SD6T'F''A=w>{XwssB§$-sm5|h uXϧ3v?nu 0`S>]c6uQqp c3mXȦ}8CG.ΕyscHpv٦$$ ط]]9I%8/(zn$ k^`$8o*K X'v#?v :u?ݘM÷σV^r9ҞׅM!pK>QtÜ|h 9q|z4._TԷr}8CGQTÜ>>穬Xϧr`$l هk3/'̾-20zFDd]ߎ?Ovo:g/gטnd^ٶl:qyܥ ]r?oCžЄ}Ic5UL^+FÃkdy~jDUAa]M56cP۫3G;smhy';LǃD@-BO=Z[HrY-v &v j=W1[ufoe͙60dS>#.8zy9ߠwfB1Xw1fx+f=^0f⼅9gNB\xNGIܬ?LXwv~ C :q{4>0`S>]c6XSj8rVsl5vϳrc[@qk+k\cg\`o@qOp-PTz< BߥcZ%8/~+.߾?j8i8{[ߘO>k8]0p< y4Ò\â۷k w%s4$8ھ{놢A놢$xNlIP 8I8I_5=٪'4דa|l>I$#OjI8I8I|t}$sk5qW+TOi'A`S>]c6$Qu8=ͪ3ϪPΆr}8CG a5V57׭` ϧn|Ȧ}8CGfΜy+s~91])8W~|*%8|B{Cqj$=9>cmSi}{z~9y(-oL?SF8C Im9Z/ɜ2,v٦۝vgv_Ց},7~ސM!p$UQvGSpK2粌iM!}=NNb̌Y- gtsQy\/T TS"s{^5.M]c6qϨx؇3C tۋas+Roc'X9B{Fϊ}nV<⬅@GaQude }/eB]2F=pܭŠ{я>I}{4dS>#qO{iM!oɉKewi=iN{3KlϾsPːkVrřb`6BoGU9aʨ#qϨ܏9 .6m1h/p3s>f!-qs~;v]"ym6X9BoGoWfhN{<, t-g;Zq?Pߖl ه3t$1N3ϩtiM!}#NM?szI\;g\X/.>۪_z=dS>#~Ugbs~<k ϲb$ؓXvVa3|@> 65fs1y…r /g?o {Ywhd K.|3s2Z!ߣM!pv\=yY6riXכy;*}j4Ƿ̑c6y3КNĺ|ۇ0Ns< 65f{B,MTZbo $Vy}8CG\qbM(AqOb'B{~6Ն!Ɲ:@ 65fZxr$k>b]kZ! SmQ]&uqyܥ ]r?o߫躁=>|ԟWU.- qAOo?<fFXcy^fZmujhΥZ/h]hJs0 O9x)W=SSZIsAGsJQ=Pj3ǬŘWK"})mn9V|h@N5fS+4ljКNs6{% 6:s+`Lgm )6')Qs ]5g}M!p$rڇZ߇Q߆l ه3t$qVWWɹ'sʜ|+ .R W_/W6B9b\WH=q|nOnuEg!B iqy9}M!p$ϟ3}-eBJpq n θFqo-g:3BI)и<ќM}8CGsϾ9|>w }>,OɼypVZRUVVڷo:uꤺtU׮]UUEEٳݻ۷+Ր!CСCȑ#ѣՍ7ިոqԄ TMMUSL1LŢfϞM ِAu[Tw-[Fu攟Uk֬Q=ڴiںu+ sNk.gw^o>~uuAu!uuQu1uquISNӧOS~:cS!Dd-!NN^  C ,A9Fig3b_new"`"!p}K s K@= p}K sAoS xŕ_ϢY = ND@,^V]-?{"P&#'^(uJr1D+ J)CF#0JB7󾻽?a{ov~V`y"Wr~*ȵSj\x"Ck(wD>oUk5D)?*4dKU4ӍDߥ3^:^5[ mFSxL#?ݺuJٳ'ՋKPUU 8LC#FШQh̘1TSSC4n80aѵ^KӧO3fЬYhΜ9@ϧ Pcc#555mFK.e˖ъ+hʕj*Zz5Y֮]K֭Gy֯_Oӆ hƍi&ڲe m۶vA;w]v+B{^{O~:D~!9r=Jǎ'Np~ͧ; y<В[`}_﶑SIfMb3 x'{9hO̿>,׿s`r^KD/w~i_c&mOYkbw19sEkSEz60k~fľ6^̩MOZڪNy~%O2dN;IXٵyy~6>v 3JQݧ{Ri'rH߇V /=VxyuNSdcs7|~$~s,3$}vy=jE8!u,:OO{jp 9dc9'pí PSqB'븆&p߳|&`pC uP[5>Os 9dC]C[}}'!d?:8^OsÝ8!uNmu8Mu8>;2N18NY!sqĚ 11Fq8qPgMb5 ƹ:1 <#13ƹ)pH |5g=]s쬳튱?!bm!՜_nGClϗSǹ1>Ob][]K 7jkܤɒ{[8_ iJd]?3aKehcK!?[>Ǧ-ϖtЖjr-8dK[k-ϖ;hМ[B]Vn%[8JlYA[e k.l 9Ė۬- j%lr[O=;c_e!^t3{^CM㧐Cl+l{mÖ-ciy[kO]C[דn$r]hC9r®x#vv:z ƐCl\5fJZl HHi;ݔ-+(+Jl93aKVu 4_v<'as:PѮĮOb^@3~k(ܮIϮvu̮SuS]7&Q+b׭';hv]he .>۫Į?OUM߫F5Gr<|\smƌ'q_g!{ܓVn{s{H#m 9'#ANsOy07ye;9 M}c8OsϷm<_/rgO rṕbqN3w|c7p>/;Orkug>_{=| 6:b?9ybO2&rgm>Ov~{ >si9]>Au{9g[oJv5FhiʙϣsU&A.6 Ow푹-1gnc{ 3|rg١g C^[O8܉!2&rgm"Ltg0[6]\&f\% <4w gncx 4{o,Б=bĊp'0l 9g#{|N`8X{8otm9ކC8|6v_'^[/pⰾ LqX_l 93tvF;3gK3hp-kHgղvkp6v`>cM!pAK48\f d>}-b2Ctw64gX#-/'vg9ؾrg"熆v9_"*8D6Ʀ@GZK>Au{xb kw0#)ξ)J9 qUJgnҔc=iq>(409M!plVa'Lfxg{V:b;?)qxo盦&oRmԔs7H}S3ٍ:xB٬Xxk6}k]W/ )$9V:==[?(5g->(!c58lӕK9hp>b pL|s~>b.6yl;6:b;?syMW欧Ͻ۲Gkp6`F\B53hp~W<MbpJ31N}M3vꥦp^;M]>G{s 睢9Ii9^>A 5 Μ#a/rgf5G1Y81 rbrA|#i?u3 6Yݟߕލ{3As }%ʇe1yP>A.6烒t$91 K ]NaN~co;>w:{nD|]B =lۨo5Q8L41c/L crgHbo rgpҀg>+KBb%{aΜso,3g7 1_)d N+#ooۛ-fA[31\:箑v9m9 j7C8|6m\'T;t:4Xk1rl>S;]9~ +|셉k3Ҷ쓚Psu;} }R\'\l༻D,[ɑip.t0qh9~h{"]!=0\E)aN|hwb>Vϻs풽?2o7K0C8bb  e>?f rR3kWK#<>s`s> B٬6  \.G17H u58(;3 6= /J;&86fY,%1 Μ/rg$|=蜟#٭t~^sҜ58iۨ5ziN3-_ԪpStvV'ȝybfk ZJBbRD%-X#z%9g\Or0c[y*kRՇ=$9眼C擖ܧgnr]o\8a`s} 6糥ko^(Ú|9eg@ ϫdx"q_rݣc/*qpB]RݗDp00^[U8,ibC +gi٣58sGJ?#.M!pň487Hoe>79n \.yF3<:6 6糥oa9蛄-&[Dw3]] \Bbz>sF w̜X[JΓ<1߯Z\@ż_{lM!<=AD3(hq?cu1_)>}Ųufڭۘ^/z/Ѹ1O+^"\ehpvqMsoEk>[;؟gogHql 9/(F̠yx.q3sF=u̜\' vI8י|-\'5>>AcN\3ƦkSVk毗{\3պ,Vzl 93tdVyn3MWNhp-ytǩmk;f˘`S!#>ν31MWNsZgԤ̓|^EBΓx48r]sizʼnz^SW-U ؉1 %XK yH+}A8O3_048OZtԖ3 uuC/rgfnNonޓk>c>?# B^=B⽩x{}crg-3ES6F:@sI:< 6糥̓,Ts=*zO}oHpf I9ÿ`S!}>[ |0JAW[b庡VG  'ޏkvHFc/Nºdļng޳T}?X~4kdfw~;?Wo挸6k#f\-um3̜Q3, N"ܝ6|.[}}ܙ٭Ɯ~;MCצoJ~)Ԛ[31C8CGluҎy^kp٦+oK48_&߳ԓ\fE_Or 6:b'q9_$Yip٦+oQg0H|^EBΓk 48r] >hQOr``T=>`T`Sa`gmRv3V)'aȕo#?B2no#͔gǦC8CGlw^=Z3QSK3hq$ksX|C8|8g%ﱪ?8_VaT8%cl' rgp%i4`|>YgsC8|1R7?#M3w|cz{J0sϨy>gnTuu/rgf7v83 j48)Y uuC/rgfn\9&皛onvl 93tĖFk <58lӕ5y"\L9n ξHWzM|#k(s n }JngsDsi?cu1$8Q'scϓr? Dw}7Z&8 c>?ÿ`S!}>swapowq6z4gz-7'8s1'8wW=sUst]*uM!pUs窿ҬU<6; p,ڛoV8߲058s7?^c4[R7T_)>jЫofL8jWEwn(ip>Ds:o·#Ԉip6n/=813t̜C8֦rv糛Ĝ9 M!plVQ!-5}jp6IRb9yO9 M!plV8pFRӗ[JuCirJ\V\3pښy*FzzJ,OG`)]$~8^}Dw9N{ZyYb>3g9cM!pMY9ÿ`S!}>[5F~crF$fY*y#g> g(<39%߱Z;9ÿ`S!}>[ Z&miNռ;€2|F3:Rw$'r'XǜC8|y^m}}ԶFrgb}X{Jߢ0䶙g_N+Շ+5 s_Ir|q;''ΒǶ7~4 ˜XF-_xͶEOL}G&n5^=uN<\{ ׳ܗܧe=ṫ{ 5qrȜ,kX`)9?*Ώhyh rn2u^\c>Y8`DwgsC8|igdθ3,2JgKu9hl 9gKn,ˉOgӜ84+}Kp(e˖o,=vR /FJ5\i|O}Ns}(zg>SuՉ z>Y9n whb>,^n!r 1w$8ߡ@\ۛ7?ݗ7gc7&΍ q؍qXzO#O+-ݺuJٳ'ՋKPUU 8LC#FШQh̘1TSSC4n80aѵ^KӧOg_YfD 4|;n<-hŊrJZj^֬YCk׮u#\N1G_]xrUk~w۵k***TΝU.]Teeڵ֭ѣ:3TUUիӧׯ ԠAԐ!CСCȑ#e] USSƏ&LjkkU]]}1c;ԼyTCC?ZpZxZdZtZlZrZjZzZvZ~ڸqڼyڶmھ}ڱcڹsڵkڽ{ڳgڻwڷoڿ:pzG}>Su!ua;__rsdN/~y­Ɇ[Tܟ/GyhZګZwE}Ej=/껝6"gVOWL?1Bm~gyчՏ6aj6~Qjng*KtAA>0lw]YJ[Iss_C4z50?OmyՙAz5KU|Ҝ̓T^0`s=x}м!kRt_g6 siZآ?淂0gt'䐠w̴*:(X۵0V:?isso< m s-1 k 9~:[3oD ׊י[~Up[ۺA7>c?1|T֌=tܒ؏d9sk;< B9c {%{N%+C2uaA5٠Z%|7co^GZӣ NqjVO;?0czQcjsͤag/cBvBtRuc3vNӅcc'i'c[!;!؉=4O;]kL cc'X\ig3?xyì}$3Us3=lN7 e$]4 hx=> <}.'dxn7: IV]{-< ~G-s}B(;jݬ Ic S>X~nh,I9]3X=ܗZo)s:3Js,P\xO;ak {%ٵBKҵ@jXX=QGvzl!\!:tݮ uExOj vC-, `9B#`ٿX; ,+}z$ϋ38{Zg6YcZ[yyg.s|m0Y籬ylVO6o0pe? B5gâyM=t|yNcr}nK2?mKBB:Ҹ?ϛ7/9- R%yƌ-|QP*1}A9:<"ha/q_k:60ók&oGayghay:ǟ˟+pL.lJ)7u{{4n {J)~H"ۦ03t1h38مMxs?c\lYr=ZXw9gy u*/` هgלMk\՝΋W n)H2{cڧ%x^{zyڴi-| CgO k<ºyx{D| {?,Ϙ_Ϯ9d?Wϗx!?WzDxt"9N :܃Ƙ ߮qʯ@\iϕseF>ߊ~leS>^sKLXPKƎ$m`Γ,I0)d]s6yzzps%y-|w>S2'!ۦ03t1O܊Ma1<)d@"vP||A"0-9:>+ [g+ ?B;#h-K.mw5zBmy~{Lqr.[#.ϟ0bvG+ }xv$Q;y{mBB:Ҙc1]o )مM)ĉ%a9ʇ2#603t1?瘇y_I¦ ~؛W4y~SO$x~u|Cl xDj{UYWa~S> >%-( =0.8485 FMicrosoft Equation 3.0 DS Equation Equation.39q%!@ =0.17Ole CompObjlnfObjInfooEquation Native =_1089977469rF > >Ole CompObjqsfObjInfot FMicrosoft Equation 3.0 DS Equation Equation.39q% r(x)=sin x 1  2 +x 2  2  x 1  2 +x 2  2Equation Native _1089794489wF>>Ole CompObjvxfqCklyCKu~:kV]\g.L!욳I%Uk8eի|Ukz՚׫$k8Ǜהr7fSGqGJYw[xB##֓3x&y~L|iH1<ǩ l9y~X7j35b+ dxDA6 \}xv$QP;y_OՈ%x_G7u)Ɖ%Xgp30ók&YW-gUZb׫ȶ5!L! YW-xp??Wbh ]ؔRv^bnd~C߳k˅rW'q?77XƯ@Ѫ?7 ԟ3Z^ϧbןɶ?j}x4֟U8}ʪW-җd^E]w iWEϋ8ǟ˟+pL ]ؔRv܍D]rǶ묺}wd!L! iKFIpܿǟ˟+pL ]ؔRv܍=}5V]rޜ$ٶ9)d#uɨǘ SDg6ԧ'G'C Zsm3l}xvK Rۑyn5bh Xw7^L/g~3Y3}xv$ wr|361)d#%KD| 233Yw7^''~ЪK,%1)d]s6ɺd\ϱV^ߞzv{S>8_'5WcgvP?lӟ5GXU[fUWu:uwb0ܼEMjhN#߁A>?ϻc^8}g5(]r}|Dm̷zu{`0ók&;CWOH$yquMޏy"nk޾j޾j޾8~DLL<3F:41Va1&ŮM1X t~oFy .&Ƙ 61=a:~0ók&Ex||!__{^ܟuJ]xcLnqX}:}_Ϯ9dAT.^]ee/kK"<_Ⱥ0.e1&Ŏ.q_c,`~S>azFzFAIԟ>p>aVO0c DAs't3裟|l}x4gcZ%xvaSJӛX,? >0>|RzAv/ί۽1omDŽ03t1凝~@_;A4~ϐ}x4aQy3t}yǧ"<)|ة7Go1N4sd۰}x41'r???Wj<)ũJxM }{t>D>8M/ZnFxmϮ9['380&2id1yʇgʇ,`0ók&38}fjHءyx=.}[}:}qbU9dAT4]}.IO ]=Կ'+WIᦛk&%x5ʟW+-TYW6` هgלM^zyB ߛyE91c&1IƜby>VPk em30ókcIM885Wo<5fsc\$q0y%u뒘_Ϯ9d]2^89ͪWޜzUe t^?Ӹ01ำ"³ Rdgc\籜wi=ioz3^S&5o$v?/YL;*O҅c*ers'jԬ$x5}jhsgί۴7g=g/tYo~',a0ókmJ2|{d>I a هgZޓ}2}mAS>l7s@qܕn 91 wrXDd7c4H?5r%<ϹH[$`ռ-V]KƋ$<,\g}xv$uUs|Z?njf}݆mБuhgʅarS>T:3l}xv$ݟ7:scgh,1clŸqϲ`~C#1nںq٬ B:rmm-k'^#ZI dRD${7z}uc/` هgלM29uk6-k+RxLٌ?aK̫>y_LXw[,03t9o(q?`??W0_-gL! qybaÈaq0]âxjۈnvԺ5+?v}-̱?uS>ڸ 1mvTEEܹҥT]vUݺuS=zPgqRzR}QS\p4h2d:t>|9rW\jjjՄ TmmS'O9LߡE9|m\jhhiP~u遲>H}СCTA5P&Dd-!NN^  C ,A;Fig4b_new"`" &g2bKF~˜V<%@=%g2bKF~˜V<AoS%x] Tŕz "@&2nG 2D8z0w@@68A\DDrt DVЀ]dѠ!g'\. jn2oJPU=ex}=s_߽_K!}"bP$+o Q>dBHQ>' qzq:vB\K+~~NdJT9NPXKdqE5ZutRGu긖4+裎Jy-RCEk~w۵k':t :u$:w,JKKEnDYY(//~z} C C#G_~+DUU0a8q'Or6m3gguuu^b…A,^X,YD,[L<⩧k֬ׯ7n6m[l[n۶m;v;wvw{|8x裏ÇEcc8r_%|:nU'b?ȿzBm輱JDzi9B=Ym#g?-;mxg;+X;Z}y%YmFoWכc!K !ګwW/tSF=3.ѩgߧXS{Tڮ|MCis3(Du+U1y '~;&w>E6:G>ZԪO\nt]S*SOzΧR-) 1Qy>o&CELZɜ}wI›5KQX~bkŨ:hwFd];^])I^VwKcuE7.c?mTxIΈWk]r$""6t߳o 32YQ,sQ@:_KDiL&&&u։l>O14#-i:jc2Oqhvy~e;&ZoύHW{NQ_w0x΍ kbC1b `kYGxC'h{[Kt j]]]!O?~J[_R6T6+k]?F9e̮] c'ו]^kWS]!u+Ov%]t-VfWV]]8[ BvVc'տ[5ص_ vmx`1@v(x`'U2m{m!;d'ے2S]ζ_oѶQ*򓘝!y$cc%]wLxZs$ l9іR߈؀t]qivmG;z?rIG? 9[Qv/|I>mKy >G:O< >Zv] >5-~IQ|xϴ!ʕ[sSaRaNf]>ދ }s鐥S,rK=wqާAc<̺sQX1;oa=Ƙ ԌtkjjZęr?~8ω~QƦ@-?׫ 9~~Lu3 6{iI8&,~8Lk8ƺgl gfZs]]]8o~<8=^gDwy&8۬@-3???gl g|5c0G[VnVl gu{ڴi8?8b1>p6٦058׷8t'8ޛ{ynkmB\åmq0`rM!l\aI^Si86J>púg)dϩg|k7ieђgےM! ?'ۗ:t?>n8lSHf/EQ?Z'OݙG7x"8<@-lm`r M!l\Iq؅w^aF낏hnb6+\*Ϋy ƴOdBʇĆ> 7zPbs_]v{8/{oml g2ޮn޼;zAbso8#.M!l53~ƫΊn Wn Бj޼y8_8㹟_tM665:lqƘ g~ +J==Yw1i3U=yܗi X̫_)dM>;i{4?x^Es)d#4^՝΋x ƴOdB{Ć>Sx)%n8ºyO>}ƟuBv䳹 .l(~LTiNU s)nkm8i.as9cm\g`2.՞Y&W܅c-?w>/%8~^29WuX/?>n8lSHu $+(V8m||Rl6,mbCv;shf63fS.8CG>8gSs)6T }\{gX~s$Ov }Jx=|X V$J X]p6l.aIvH/`tź5'! 3< 9OM!l|͓w#'T|r̦]p|̓$4uˌ30.T4٦6ﮡT˓Sϓܦl g,)>#]o9 )6T>9Fx%s+8#·M!l'h;Lm y~"{108Ms[?a{cL~69 M!l\淓r޳|W~7'1Bv:1o˜]6Ob}s>Dˇ헛χ6l g|XҺ9]o _dB3>ⰷ8XJ-yD,dg l{l!=_)dM> 3}4җq]>J3[^kq]xp={’o2hoG;.d :1ɜǟJ`L&R? _͜S<[ >Ls)d#Iu7sOy o|lM!pju_sټkyfuۙ1Bv:򱎰f͚D%|& IgmMMs[1&o0#Y8!k%8`,;Nl^G!_)dM>:Bɜ8jy\-}༓u8}P6#/ &eaĉ-*so3ίʷ%8_em3x;l gS. Gp6٦l8xe)&q~Y'se]ϲ>Ⱇyju `rBv䳹c-"Ǐ8(_}".gquy:M!l\1[8Oooo:l gvRM>86Ʀ@-;l163}Kcl_, Бue˖%|):c>p6٦-mr7>1ySc<'_Ǡ- Z9oSsCdBy̙-⼁yn>Rs%7bQޭիu*l g^ğ1!ۛq^'{Kph8c>@;N?qZWU/ &e*??͜8݌n Yw?̎K>~CK/ &e]2^N^Z^EZ)d#U 8/e3xM)>@܍|y«kJ{W[I8]p6qlp=*IrSyTPKjnkmjC@-ϰ1&Ux;$. M!l\%U+۾իV׫hnck$d#UI Ss)Wgm &w^8A3Ώˇ%8g~O;N x]X\]p6l.Iyϥm)T&+R]?/q՟ugl g=<ϗG$ež^;({V!ן_)dM>s^ĜgsT E9zf>pzq"f C?/Ҟsh9>&vszUbV׫hnl gzURAx ġ}lM!opzy'lҥIp+8?ĺv\G! C?ÿ`S.8|6_ V^(|W3tV~:^|6 v@v:^'yy:*}.1J|lM!on|a >%{K Бuɤ<OSsyz6Կms=0Vkuz9d=`S.8CG>%9ǟJ`L8lSH}69sA\'# :]>s{gC3]p6l. 9B?ϗUןinU1! XNsc?>%>p6٦mrt>2'\\b_]Gy.q C?ÿ`S.8|6x.涔xqK.+]k߾ǽT3!`S.8|6}Iq, }YL[y.q=uY+U\l֫$mŞ G6Gm܍^Upam;.Yn:$ 6삳g.99C\;D]rV%k5z;f.YSSz ġ}<=G9m8|4.4C d#Iy*A3SSYwێȇMq?%uuIl g.'ܖr2S8Oam܍ C?ÿ`S.8|67\o:zrdxD]H_4NiE\+Z&:G=S<T;Fh៛b׋b];b61E4D<^Q'Ύo{>$x>o$ugeyG =<'^{t|A"̺v_8of:l gw )1!ņ19<G:8܆l g&ig0g`\%6ĕl&ƪk}bpϺ="M 99i9M!l\fϞ"WJWUWWbQS}VS!ה_)dM>˚rR5O|QzϗMlk tu sy ƴOdB̔MGu\RzMlenR:9]fsڴi-|ƅ2>wDqsM!lGF_Ul턘M! ٸg5$Ss&ҳelJ}U,ΐ}L3vkd{֫ Cյl.{up]՟/_)d#7sq$ߔa t|X1&mr9gC3`S.8|6}Iq ^S)1q$Hy.z| Mc _)dM>>xM:/o_H[k xBv䳹3Hʇ 5u }!8#,~fʮG==E?6gsYNWsϗMuTC>p5eT!%17 &e]2iiZjl׫hnd#z%AVD_.8CG>֫K?y tOdB3Yc|ԫ 1q5+8úXq:6 6삳gsNZ+$$s\*=]p|̇%=8vƹ]"s]?ņ>9^U0{B`Q}4l g>?W6ǟJ`\陝&R]҆'飮FfW}C}07 Б}CIkX_/0&slzūIlz|={qdV%_ Б$|cZ}5G}ǺMǺe%]Кz?i tOB8giM)ϹƘ g;OR{'s\!s.~. X> X]p6l.aIiøD'5;~54z`NMTܩ 8V3巁um 6삳g5}k/(>Y)gZWHɹ؟?VGO>+VB3]cp?gSf1Av08:6 6삳gsN"2P_翨WY/>@6ap;o/ &e~;~#=io=iLi'2=g`Igm )O҉1Av޳#oۘl g-ݞΊX0R&RM|XgT}buΊQg\ؿmW[?w|L/-F{s04|رϔ=rsM!lZrYHU{9)^q^yOy1Bv:1nӸOfc8lSH}79|QFz7s>>^o/ &e~;iݦuNZ>a6ki.Z>ꨗH,Ou6< Bv:1O9?5Vƴ~dB܍M <,:qb_]=[8SX`S.8|6_9x_4Evæ]p|Ijc@pk(Yn|*`>U:d^]p6l!;ʦ>>'%r%>p5}CvlL X̟17 &%NU3O+zW|M_o>֟m^<{ƫuȼ Bv䳹UI6@WmS|-G*6\723b X̫i8o6Gɫq8[(%^pXX̫07 &͆ Wެm"ǡ~9BС 潔+Z&:;Xo(eN{ZbJdqґd:o$ugi,eg%5RT}V=垃|>uuu-⼖1l,VZ]511Z.됹 6삳gs}bM0!&ޤΆcn)d#h8m86TS>?X\Ls8fS.8CG>&+g`L1M) n/Wu{:]p|\g`LM) 3^]E3Upkcg:sM!l|ݟ30~Ǽg磭XC1Ob3ֈ[kIO,߶kNtAtItYnݺ2Q^^.N?tQQQ!z%#'  (, "*F).rqW*1a1qDQ]]-&O,naQSSC P͆NQ__O .XxXdXl.z>3$֯_/6n(6m$l"n*m&v!v)v%v-#'/8 >qAGÇFO`DdJ  c &A?2Kg8=`ؾ'jM`!g8=`ؾjxcdd`` @c112BYL%bpu .| z- `+KRsA<.,BF0( J-fDdJ  c &A?2Kg8=`ؾ'j`!g8=`ؾjxcdd`` @c112BYL%bpuFig5_new"`"<% ri%0 goq% @=% ri%0 goqAoS$x] Uu>羁 B,XtB Vq@!B2ZZHԁ`ŸVu5E Zb+b D!LϹsd9o=7k]x}߾oBOFQ8Z!]&Da_B ').D[ޟʄD~wBUɌV9FTJ_1SuRGu::c54]u9WdȌ--++Ccǎs΢K.[nGwޢo߾/ C Æ #Fbbرb'O555VL>]̘1C̙3G̛7O,X@,^XՉzdl2|rb j*zjfSO~Z<3bӦMb֭bbΝb׮]bb޽b߾}bСCȑ#G8v8~8q_?5:AMY*?yylt,a:~Ϻ/T-gsmxny*hze?sQcSNz^W>M7}<w;useXd)JVC*r5i~!Fm =&Ӗ!{ WzSܔ5D_TRE|>טcBN'wLI꠵:z yX COٸ'vKnߊJOyurH;_PzZ^GykY3cq:qoV̽9z\nQU|'xh_m45_Sd]cseNr}Ch-umNdW/=EjۜD7鯈鯈dsRsۗٷ!s2YI,$c~KyR,9&Z t։:Dt6g'՘~+HscZyBv'-8wԘ^w|EMl]I:&:';γrE|9vfg|vi#|iK_rsgﭶ}ow>5#}G>]ZA5:lGd:yřr wV1 UYB84n7IB# C_)dלުI8qY: }[8l| naH>pu6Έ `SȮmZ瀳McTm=<\{nuh*?=Øn8lSLf/i2,b/ec4/8_κ[g4y$cur$-!/ &-d]{Zvi>8>fSNusZv)3s9i6Ŕ }|!C.c/I>pu^#=Fu6 6삳g oϘ1YJ{ƹOCbs8#.M!l53~ƫ΍fϫhnc63tW%p6٦6+CqDbsW]{%>sW!gl g?/_Y;wOe;GSGTάq0@-=|<!gl gϦ?wXB?w? ıS92s%b18jx.\pFBဳMl?yƟˣ[s?s֛ t?'aWd0X&S&6έWstmustioĆ>6<^Z> Xk}cqak֬i?k\"s|}'9.#9% tq&'a t|%w\ 0 qÀu0l g2ob2%bn85~XKZsqįhy`rBvi͓гϓ|$w' Zƽq8c}lM1%m8< } H<{rVy۬M! i̓$qHכ8c}lM1%m8|Q:ia4b63t;#]1Sgml|<'(&~q~G*sjpFBvT޲>oRJoC7lCyC M4C75pD6~74:6 6삳g Nop޳\{!>Is#fS.8CGI Fs)d#$|9]o _db3>9Xi$ub?Ć>$=uyOl gLE?q-],߷{#~#(=:U'{%_cq\6;&Ww.d :Ҙw4|㼒qo|lM1spj9>M4=1Bv:Ҙ_N5s|d0&~gmodžS5?[:B~2!cy^]ϲ><iX\G]p6l!Iys⸿d_Jph8aݭ3x;l gS. z9ϛZEY|V)d#"0U30&~gm?ˆ_b⥌ rDXw?ˎc~Cc/ &-d#)#rDԟe|Ἁq1uu l gCZ~{|~5fS.8CGp6&_84y7c:sm_ɩ1M? gHc#??u\'`L<&SMGc s״:&:-;.8CGIy ds΁1qh8lSL6G>l#>\ n獥>@'^ǽ_Wmԫ6jUS\J8o%87Rooq\WU/ &-d*?cCvy]Cy.qzq:+Z]X\]p6l!Iu{zU^ՠ <ԫR\Jʓ40ģ}Ptݴe˖5pA&{ n'Ywk >Bv|'5pݤq~BekU#^UO٦6d t,c[sZE6~O0&gmۆU2O=ܙ$eݥ~O;|/{V!ן_)dM>2>՟w?ލ! i?'m7hU4+bd#y) p/;|w2$[N[ ;Ywk Bvٔ o1$EƹN~QCuz8>:ZBv䳅3Hʓ|skʉןinc63tl y" )_gmO&G#O-2V3΋j\R_]GyQ C?ÿ`S.8|$^u+sۇW$r%>puzr7>[yk}:>l g Wg@.q=yG]{_)dM>[ށcB o`.oUr\6YwkLs)dg'qg1xgm+g{cL˼?/EOkO{DkYw6 45p80`rM!lBaI8;8wdOy.4*|j9.`r Bv䳅q%a97>_MP zm&1:X3gN8O3'W%8'iɶ> <[`r9k<p5Zg 6삳8>>Yug kvF̦]p>Yýuy2s\gS>j"꩒9,.c/eCew/8_ĺKyO!1&^/q u}/ &-dARsg|U!/8_ƺK0!b{cL=Q'!`S.8|$IyR.cl g ?*^Sqܗɞ{HGΈ)KYqO: qB֟UWk\ & $z|k-c˔/. CKbn)dM>[ȺdRsKfhQ2|斉 Zj(cPիzʦf` t^t?egr=gmfq? gʊ4M! %+س*Dڤ+ 4~ݦV#]::FY߳ұ3u|X?njaav)d#իW'<ں C^_)dM>[u;^U1UZ`\| mm̦]p4&}WQkW3s96T>IrʵZP_|}C4c63t7ԗkyƴ~dbz~Mqx8>p6qbz>MϚ^՛%zUo!zͭC̦]p4֫pnR vZt;}ɟ~D3xn0>]s="u|x{h&[cPٵghy)dg>[??&ϟO„M! iIxd0}&S]&c>.Ihof/4:Ov:3s9i6<5nwط0Zj+p~q=uyOl g2?qPW!'2Jc~>yUR~'X5+8,P3 gc=:FRģ+}L]Uk9{cL& 369 M!lB淓(rtggIԬ|LXT{0'9 M!lB=g1ҙZ>a6{i.Б|X7d0ڤM)J}rޭB{{l gHc39mygr>gmOߦ&#Oґ׍vG^#玆b߶ɹ gʅ}C{&ZBˇlesM! i̇%4`2-M)&^co֎ ɞ)3]cf:#y'9 M!lB=8_A˓Кzf4:Ҙ'I;M4! iϋ/NyJcZ}lM1V|gzVl{3s%r>p5J}v|L9`r> sM!lBÒⰗ8xK4/s4sb63t?'1qާ^!'}7dq?q Iuyl g2ORWW,8P+]UxMyW`S.8|6Ȏ1dI\qR?]G`W}།uvG}Pu6 6삳g o'/Xl:]p4۟u\l g2\]R|3K.]?qzyƟuBv䳅ϟuLxٴSنӏT VGXDDY( ]0nnu##=>}_Dg4c~a\-kcqܵO'ߐ]9H#]dI8g hƳ֫W>p^/ ,4lxx={cbl g2&N؞Xi:z]p֌30&SMɆ؟a?? ߟin1Bv:B۟q?=Øbn8lSLc~Lcm 63tq8cZ}lM1&m8uv^7jm%1Bv:B[_ g 63t~8c}lM1Il8qغ=Og\ī|L3~`S.8|635WϘl ghٵFܒ]cO%R<$eee\tAtQtYtEtMC}٢B[W_\|bРAbȐ!bذabĈZ=Z;V?^\wubɢF֊ӧ3 bq!Ay@l2.5b j*z={jLiW6m[n۷o;wvw{Go8zxűcǩ>fkDd Hph  s *A?? ?3"`?2R^Bsfڪ.`!&^Bsfڪ@ xڥTK@~flY\Z"B(C/еPktUv=?{(&,gM$& }޼.@=\.3a$ }>nZ.D,yGb%gZ4f\k-e^ PkQn6lgNAˌt7S;' i)RڡOG"mQ|B뷕?b]olR"iuWsAB2By.E ^V`z|%6Ac;=Eodm{Y~||@1'Yí\I_ οy{"pu^5%jD+!U =?N*s38r}m;zydysGL[oWYŽDs|ғ~r]wqŧnpy|Q״jbV#yn#GDd 4uP  S A)? "`2,ֵQ6/]'y`!ֵQ6/]'y4@Yxcdd`` @c112BYL%bpu .| z- `+KRsA<.,BF0( J-fDd 4uB  S A)? 2,ֵQ6/]'y`!ֵQ6/]'y4@Yxcdd`` @c112BYL%bpu .| z- `+KRsA<.,BF0( J-fDd aB  S A)? 2,ֵQ6/]'y`!ֵQ6/]'y4@Yxcdd`` @c112BYL%bpu .| z- `+KRsA<.,BF0( J-fDd9J  c &A?2Kg8=`ؾ'jX`!g8=`ؾjxcdd`` @c112BYL%bpu .| z- `+KRsA<.,BF0( J-fDd9J  c &A?2Kg8=`ؾ'j`!g8=`ؾjxcdd`` @c112BYL%bpu .| z- `+KRsA<.,BF0( J-f;Dd-!NN^  C ,A@Fig6a_new"      !"#$%&'()*+,-<0132465789;:|~=>?@ABCDEFGHIJKLMNOPQRSTUVWXYZ[\]^_`abcdefghijklmnopqrstuvwxyz{}`"f:U?HV@ vB:@=::U?HV@ vAoS:xݿuv}de 1$,!DM+ osGn{YUc5>wgQUO?||}|G/W?o_?;ǿ;~?>>?ew7ݿ㟿ۯgɯ|\|?ɿ7_+~}_ǿ~7~__}I??? ͵猭\=;!y}d% [`u5.sxN5r5d_!84Ïx}uon๹_j 2 Q?`@ǵ@@GrD_{ Vס=zg{}s~ "?q{?Z!e~xF h#m<#{@7:?< t|y\<{m21G '`n! qlϦ0 ~ `L z>gf=eP`n~WP+x?0$w?#is=z_sg|nCsxs&|mi3!k;}~7wMu#Gqx?/]>]sж}07G{#}2~>sSsvX?%g|g5ܥ]Ndd3`Y}UnkX:hv@|P"r=7ٳ~]f_ý c2~D難SϜ+Xdr]ì:Z1}PyJ֟͘cYxt-mu#,VaKC7~/+32VuY9s,r{}NdS}s5gT[zב=?S^rѯ-so&W蚄ټQ߰Z7;2#s@DoY{ȱ곌sWj/)YJ&A7OnfʰgIvS2cծQ+fbU2kĿŎdG8#8*Owy^Ϩ%g6׌mŻu03oĻ]kx?<'f ՚쵿#3XZs;defMSߠo||}vJFڬlteӨ!gt谞9B_@&ze^ "Gedw+پjz~7z.*k"^:!@vk3kWF^Cv_ȬɬתsV?o|Y7ZqVle̛Yo~Iuf5ьuNi3k3+ЩȏVt^3""W1 ܠ`̗uv=S ص'Nv̬7uk+3ysV?^_`|5yXvMu'֟{ٙ0sO̱cݳy]{Z+`0O3xveA};V֚pe;gȬjRfe']o}=%Yr̜+cG7uY GsWkYכ9]#)?g0';vM1]Q֞UٿS60'կG<{cMYezUkh+kwsBOs k bt}$oWiߚzwJߨ8wG}~u&Quzk3M] ` Ul*9G ~5>zw޼;fghe]sևwG+OwsrOXjnԬ|@F?S]kZɿ"j9xv7"o^ѿsf=3wWGVtY}9' <)ONWYU5྽oߵ7K{}G~řdQI^ܽfhm{|Ynk{벯K~vGޮgw^oRn`W#}3p~U n:#A ^Yש?ѩNngLy~b~;d~{vPt'ͮy8Q3{3%`v,CM'ft3{CkJwTW}awpZs}}fl𼆳F?cџɝqat6l]TG;Wey?Fj?buT9;Z+=:L|]?:!;P}s)D<.F{ۣЈ96vC;:/d1]k2uy٬8s b?Y;gvow-rxuw8kP}סO^jO>_/"k{:o~Jy#9 zߞrvWc3{ew_/?y<6=soNDԢ4G/ӕQ[O[2{d9'9ጁk_6n+YD2oo:2"OWGV/F;wyʞ#nW>c!b^Jߠb  #[ӻξYq`{\Uh4o Gʿ=>֦FN9k2/FfxM;w[okw~_kP@skիǎ1gdw쉘û[_SUq'׵]{|s5UeygVk=RV=Ğ:_9=z^z @VWjX}~D13+)cY}{ ѩ:Ӻwc.;lFY?@}jqswGRQ5:'ZʝA*iW39񓑽gyيȳeՀT vCζsHFdeG/g##׻=W_V} k|ϋr9 '>x~G:\ߵ݉seoWw_nm~9`=}2rY3eOZ#W|[1zQ\Zda{58(#ѿe_}f {wvViρ=-^ qX|gfs;c<2S^mg櫚3պ'(ܷk *>|Is?gיsx{kښ}Ӳz9OZ;ܭ0{ s;:_ k eܯ_ݳZδ;\@ ߩ'km\uTMi7ksOvۊȵ=1NSZ5פs_9kb5)efݧy9<mM]z>Dߊ>,~ sYs Z:oҥ^3sv{G쁐&mͺz@rյWjN/fkQ+wBOWH(2ru3};}{>?WsgL#k9]kA22gHڟ}X0==;{*3uYq8 GgU{F}gu'4P3k\v?gKankQuπ+Ϋ{55[Qߟ,x?ye5GokxU;\+؛gy+({^y亚 gNkԪj5Fk߈>@Zc'j?߳g꺦Ƚvܥ0><+Gwo> յ\Wך16dcE.җVcu_=*'Fjr\yfgGxTt \Noؙwwjg͈JWۮ1ߕ}6|̎>?koD|WgWLn:zu}\Oܞ<#5gYmgWǫ3"cDrVc:~XFww4ڟ|yּzn7rL?q}sJ:M^=8re/>@3ߎ@׳26Gd+W\>s3gBr 939n|>9곰뼃'ųwY:~>rݟѿ;j cﶫI?nvsfnzuu@Oz+uG}O޽j[Meꞡ;3;ڑfΙgN9.?3̾\㙭Xk>2ӱn7c-=~;u&@vF7Uٶig=_}3kM+*zT@;BZ<7 0;f`_zFVgU3>5dɼ ܏g7D?3f#s{cf=ggxo麟Z;?kmh?b4Q7|N\y"k~AߛC0Fy^ٻk|u{؛2rzQwڟ,8gj3VGzj`f d3Kf2սs2V/3r>V{m^Oy*Wk>*gv ή?v0lߙ ܱgv.{o^B:]{Td'|eMūsZ bdWf_PW7{8)f}60rWV8oP\ͳq 5;kLjK3_sO8w dGG{t\α s#vu!W䬙s1_9e cMloV~Sv/p[u?.=fj8g5ͮ/kJ~TԘW?TԒG׸f}D]~ώ_1weUo$_]#2ϽIg~.rL;,}{asgre|]wd9y`9f&zg|^ud]@TnR<%Gɕqp3:e}1k#Е՜A\7!o#䊪@w}ѵDt7wcc|̫=#G>})*29W|]}Bf)ٳJO9NYze9LVgG{FzQ{f|vmeNSs}iktX+."gSլ.g)Dt:[KRun&c,2wY+ʱ9|^σ>@KrNpjn~kKzz|dND`f?@N?36!OVMn5w tGKGYd=r~8!5:ߵ~ڟ33VCw'7g}FW?{|Z9>}\U@D%+Gg.fw>w?[7>Atzx:9kFձ3x? eg"=ijw@fНϬ ~i?3WÙʽy޵ UcڷV:+Wwcɧsyf.3r~u?pp^{TNmSjVk2ٿz3xpϛW]cYN>>cB'~Z5Nk}wk\d;h97puנVڟs&E{~Q%wyO^?fTG}^wC1cj>X}N߫7#?֑ȼ W W2={O3kw92+쿹+g,#@ŵzV`4u'NyQjow .Nfgrnv_$vg%Geսfj2|Ƹi>@eMM@{VKqgDKYYȬs{k`d6_/_ٿz.@ZRKi2#eeЪ?kqGاy>{y:2֮h\{dȱʵn=#_w;̞֟iA JV9vbjs~׎_hoЕ55@;ϷЕ{D X 9J_=jZ cKO5WZwk߀>1'\KklȔ;ctZtvbz.54OOؿY h^>m`YL2cE-vu?}[ϕ53}X z֕Eř5@wZv^?o`P37<`L~krkVVZ_Usj?^hp/du9xG\]cg~\2Sjl+wYY~ry2s';ܛgnNý|sFk~6g u@Oy|y;+k3i{Jd6VN^\g{;kR' ½eG:7v崱}=q?W53-d~Z~W S*s6OOmH܋5GIzjlݯqP@w7g9Lܬ5uIO}Y=ܵ\x9=?g&0 ]WgkN^3zvІBֽU9u`GK:nt_;e*P@fۯ;=q@^Udϩd*ϕ9g2 wg&8?+;Ww}|ٷ9³@sNg WSW ;=KxJ`$g}&pyQ'fz9Pm 댁;'c.ؓORF4{3?= ??cvgoZ,}fW9cٿj3NM0QsR<>ؿg( ztY[?mo{??swz_Kڟs 9cSj"Π@TG2r3 kOܵ9ﭪS7|Y?# PμU:찻g( N괯jFg9뀣y??:pz߯1kxεWW~{u){h+,w}(9}Ns{Y?@<}W8r[U>L?@̯8s z|j4/كU7G ߪm]T>z]}o.s#=/cl>@gvom79:WSc`jV3e y`d1gUWvw7:{PXu;kϓ#^)}**ʨgVju&ms^s3Y9;k>ĕ}An9Ddnjʺ՟^;dPt#"}&s5֎dtɪG#@>@Xz^Q{hm+NDShT;em;fsHƍæ>`/,^?c\Ͼre+D22z~0T3=P}vgvf!u?Um{2u-cUc<s4 Ϫyٱ3Gj|7ؽxV=vhgX#QѧAý:jZFƧ3sJ]d:;`6fgKϜ:|:^ULy=gE]v#M:X9ved2ܟ2v]߳ @ Ekj2wٝc?27;])k/S6ܫޣ5Ls{>@߳ @ 'sg M>@ygmhgm# ۙeWm,ue#WೠW1G+~_;cW>mlܡ1:j%>_`J<@EW_W|s57dF=>/dΟd1Z?z'@>cmq?6ܟO[jy"hg/M>[Y# O,fTOn?=;\@<糹PYoܟ>@UͿM|vY<ӳu]}̚ @ӯݫAW~^cr6_>}7#쿯!@~}f*rwm-|ӣrgB@ɾk~%<mEdVxFtO Жf3ܟ _x@6xv[=T3.x {@g[73+?j}{X?08q/r?X? = k_Ock=[0p5OLmx3{lS}hܭ?Oh{@3 h3h=?=x@|y@S- _ '\;x]y?W{&?1@k *=Ay5@ 7'+?p׬נ>8;`~@ /2?}}kw Ww>_{~5W_{5oWߟw3ks_geW_3Zc*+癟qw_;ZG?37dկzF_@?2F;ڤ>GѯͦWg_OwLC?]mKD-/pwS#_绚w>?$[n-p:?:1rݮX;}'UW~ky#g]٣g#hpRcru˧ {&wHHT||7Oo~?]}3o9O?||ÿ_ǯ>?w>>Ǐ񫏏?w>>>?7~og͟oǏ??o~ko~7_^+~?߿/O??o׿oOww>; y,z89Y1 |"?z A2?}#\L^gh{<#{@ȿ7^s}̀ t|y\z6 ܣ07y`mzrMcl~+sƤP933Ͻ_`\=>Kxz^9@Wp~7>0$wpwFG={wQ=*\#:G=`S#sFF~q;g3ecdX`9~^{n~ѯ py5[wfDI s/~ܯwdlވ9W dGϺ?z0ޏzfO-z@ Q_gكS٧3;ˀ~ՙ]O?Zd U;sc9}S끑=6#쿲 5/quD=>z_eWs{G?:Rkk)^YY7{L5С?w$?gdBY z0;p +.+_ku+yCW8 5ܿ'@ DgѾ? {OE?33`"̯Z?~נ{7X; Wu>tݑ#ΨQ}}=U獩"={Gٿ3|^{@{ox?\ǟY[)?uxֺu-5kOϕ1n}n澞_;dr&5#ZxAex?fsg-^c\=0Y ]cֿocgz]ͳ?`kAר5:;+d=?o?wDf?[{g3}Z]Q~\w2?Q;Y}>3c9s*k}&AWEj j]&2_;dW뫾{zf`|3;3g?~Z'@=g^QWtZWcn8K~Qχ>:?"e}M>:'1T=}@5}>2̯lܷj~!8W~Nܠ0JݏƑG\@4Z}m5omY^=8Q}H㾙=c+|Y x?`bެ~c5@3q4_6j0_ ?칅~DWrܿ?!{]ol뜩5tT|s:kb=ot@Ec WlGn<7;f1gU>98wMoDuUoqY%ܪ"3?gs*ο6'~s̍}??7Wf{\u}qv6k{Ͽn`F*[#5HܯWNe:Αyuw5E7T S&f!r8xZUU PNEkw yѳ<*k߈=={%g윗*12~~A:LPtv:~ɵU~wnWUS?j>\~fShkZ7տkz|>1_![콽>QAν4#Q?W0үd]LYk։;Y]2Dק:g*WG9v!ٙ;{jb#X3i?:Zkf78w̿"FAaft_FuѪWt5;߿Zbjwؗ3L_3ca_.=5@E8_Pڏ绯OG])wAgAٛ_}fh-P]fYqZtC={OZO||~rf9dwYjP] xB+޻~ݏeaaCMFK5|ҙE]ֳigu֜@FzO}=~]e&zr^3*sv>;BO[ zY[]=3w#?Yi,,+uJ@u^xO*;?/g}r0~[eաޜ^~s@Omݫz.g]eq쿣~O>s,J7SWI,GVߋ2l.zûLl߽߭K} #={+53}BϹ΢'빫83#V_OG~׹}ѳ ӯ׿C;w}Z'1V}^EEt[:WҬ ywMSEg. s3mw8s y3rf9 ׭^yG}]fdO_r?;˰.kʟ}jWgr^uhv~i:{}n>{ck'U\ЕΦXOWe+c\]̐?Wvci?bn$X?'dgyƞw;Ǧwid_|٣?Syc{3y^<p{AvU9p>=r}b#OYӛ5^uwy׭C_9K0jq={X{{A;:={N"_5w8»C{|^ϏjTQns_#j_zv̷g89[jz9WY{#97|c +F;C]TsU9euYc7ﳻ]#<{0 q~ {+DdSjNs;3֒T:lq쿢7&f\w|]7jpQʘA9SuG%;Ȥ1zOAA]6=C(>;zg 5I{G5NfQd&=?b%hW=8z>>]S dͬF}'Wtc}}7'9T][#" UbﰾrYޚ{z:8s_5WV'8iust^Ck^G~Yr =?sY?zԕ#z˹2^up:9nsQ';˿gkx>d\'N]йӚ;{ɨ%V{;w"k9?RwXruN5#wv>_Ƚ2&jONgżvd@gϺWW}w[t#e~o̱ٿ2dz/c~WFQ to8򵽺]޻]vʸ]X?,]Yc+x^K3+cߠ5k_G|ctoOz23NbQ?TIc{tگ CVϸ\]=wqݝgޟf82'zk:o= ֐Et]TdNCO" D23G'ksˀ^s\l:ku]]>ף{2rMdsH*{Ke#sd~ȳ2Ow贇~7#Xqc+gvoqkF9q'+Zڎ'~}8ׯڡ.Y\^UfRtX[\cϯ沮߻z=izʘJ 6R;DeP7[Y tZu>nųYZ@XzU^ܑzU{y^}F'SS{G2k 0?]%2V/޳߽JMڎ_2~83ݷ i5@d̍$YJUcF+sguƬ~IӾρ; ;jsvq6lvLٽj;򌬝GZ=W?#~{g]T SWU4g^/Z]s/D9\go?%{^6yYdYuϟhz~=JW{P"]xDrRT2>3QUw +s3ukf׍Fg?^q&IFy]9z=ό0~?b|Gݽe0V ʜdtauZ۟=`&Gw9kK7w 땽OExz¬\]ԙ@vM^ݫb_c5]6;goG\z{<-~Fez#{|j ыx8g 7Ǯq+:љ{"EyF #{E{VO<_1sQwY϶24gq{e̽]"F!"k :Jznk;G73:޾k"+~t_iWքwznN\}F_o_N5j~' 3{V4i k3Pwɫ"?:}}'qwngyKt&#Sy@FNoѕ{dG5#_97bMTϝ=1cy^sV;b̞<{9=n2zvѕk՞?7:+@gkV>]zTTl2J +6cq>'2eJk{;ihƯȃg\u]*3]z}8)?9?|w?uhG7*eWG窓7~w\9JCWOtg~cW_Xt?83W퍘vt+sΜzϋ~W>{d.Yne}g>w>|ʭ:=z2~D~.E,2/̉>_gt؃z嬝r9s21礹ȳ&xwWH>ȼlN!_?k}qqޟu#=k;7&+wfۥWǫN]1&WyZOy+>Ӎ=4k]/d̫?u kMl_Y_dޛNϸ-Z5?{qrpgt3;/3z6u zL7{Ǫ7+k}=f쟽i'@<@D'j,{Z᫽>>PqVy})s/LU\3kxz7jʜ2~ݳs+]zPϟ~3ZWܣ{2}mz[=+s+נv+g֍_=wl'Evw y^@9{P;jޟu5+kfNKدu:'pswUX_aϣxVY}_GzW|vp𻱷X+[ WM5Qc+ʹ&#?wy3wwe_OZI+3V3OsWy=WN^ 0Iu{t:]}7W 9_Uݵ pZ=yU\O@F4wWf>gud[ zmʈ5@>{ϙ9}W>x<d\V9̞衉]ϸ>#a:^Y?f3үwھ> Lc_ 0:FZjd^o{k{^xu<}=_ti܌=@>s%+}~Ȭ^g=_T){~.d;kϲƾ>w{ך]=0&2<Gf]vW:ٿsn 0g{32d{Y͜51=k2P=f}@d}~z޳~pOsw_U~w#lW{~V?r7dّ#Y/wFȜQv\7 c⽪g #c3kȔ<Otk髵Qsw~R= eyst?~JLOt1mx3%w^\'f_^5z*ϭ{]O홋>˘?L>g|'9eg5ve]gԹ_Zk{=@rn]2>ns{*!ؙ+pN,\vy2{wؑ3s2] \;wpõSwU~:?s}L3OZa\K3mMl^ϼ6gsu}u}332g1P z_[k\aO̾N~=9Z>{~?SӥgvOﯕӳ??Ͼ^w^svk?#};wm;?>cr<Ǐg(\_2;]뚣#KxgggZ`wP5?9yu=gz8j{*REI|jꞠ]?쾖gj뀳{mOzUxgg( yWkO<6N^/Z\0[^N@fj]:ޟ׳ @~y`U=@y]zؿqdfήFOu-3+?Y wj5_=+Geou0g}2ϊ@9+T1guWgf9 ޢ5Qb)0[ 0Ӣf}jtu?#Y w3f3J Ny`]z") XK'_WHn{P_&|f?l}@W{sf3_y^Q98=וīg wͻގ+Q};63?q%sN3 4p%_Ɩ3lEޟ3gw>g"DZ+kwyW@@9sc3{Yg_G \αS{]^q rSz*zWgOs|Ogݵge0Μck?x]^0pjy:d~\syiyޟs@snܱؿ{= ``k܉3=Cܻ@ \W@a>P=v]o??O9.?W[<'gv}5lumf YM𕈟U?{"\<*Gwg~Q=>K3{ww?5z~wc֕=<;̌埜pgvڡ:wwocku?`fٯ{i_{D_*Y<{nΪK+p{5Ξ==O?zir Cu?{*21{ܱ8y@v=4p@vo]z{%w2] ݑwDg,U[@,;{qks߃9@fͨ-vB=kWOʳyyGiv<uk%|x25=@5Y^Co PiY- yx9yȤxuq@{;;SwzݑL3WxoX͒;_ku3= cߝD UTp_p{g:칇ȵcGݏܳ^6um=?[kf\ݏؿ+{kNuVyHw_p{Ot P++>_wpOõ<gW@õA:}8gf@wpõ® #GuәsoLk5h1cs dd8\K?ӧ3+\2v/ bW .?+sؿNu+Ted2\ucG18/{Gkb9yݯ']ud ).\9}@?U_@ӵ{9=td.Pp-w:DP1/|k@UK*>wk] 3~c>|?)W.U S:>k}krz>rsק{0[WYc_=[ܣyߥѿ7gfdlட 3_Gg1#gs'?=;{pg3r`dξc|kqx+{:欧fdgy=cV|캀|l:]3}5e͌eO ?>Oz2{fF^Y=wL"ԘO~#\Ygf|3@^ bϧ,+*j'~v<#{Gf =ܻ'*W~{9xfo p5 _y:`?I߽@;3' ߽p}WWzݛܫLg>ǽ >ܿAƃ36?| ?@{tu]Q6Z@d4\w~z\*w} c`~ ~ck0?k`~ `l S50?k3[ܻP{ PjhܭP=y3 _@g4opcp R 9d/o@^|yP/z|>w=C^+Ț'71Ȫ>Pȃ '#/K@=w ȯz?z a@=gP}5p~=L_G=^{2?D|W}>ʿN<;ˑgkW9J Y6WU-pR2n?[C:ݘc&?ՙY]O+;u JQWjd xOrwE]#?:d?g"!ww@F3jؓ@d?`y"z}"յJWk veoOfʙ\Wݽ{3Wr\qv3#rxVUen@׈?X˓WK"oj{ bW+y^~c?>>~_}'7?oo?Dd aB  S A)? 2,ֵQ6/]'yu`!ֵQ6/]'y4@Yxcdd`` @c112BYL%bpu .| z- `+KRsA<.,BF0( J-fDdJ  c &A?2Kg8=`ؾ'jRw`!g8=`ؾjxcdd`` @c112BYL%bpu .| z- `+KRsA<.,BF0( J-fDdX  c &A?"`2Kg8=`ؾ'j~`!g8=`ؾjxcdd`` @c112BYL%bpu>Ole  FMicrosoft Equation 3.0 DS Equation Equation.39q% XF2 x FMicrosoft Equation 3.0 DS Equation Equation.39qCompObj{}fObjInfo~Equation Native )_1075270332F>>Ole CompObjfObjInfoEquation Native F*< [0,1] 5 FMicrosoft Equation 3.0 DS Equation Equation.39qzh9 r(x)=10_1075619903F > >Ole CompObjfObjInfoEquation Native _1074685141RF > >Ole CompObjfsin(x)/xx"["10,10] FMicrosoft Equation 3.0 DS Equation Equation.39qR:d p()=12exp("||)ObjInfoEquation Native n_1075012494MF > >Ole  FMicrosoft Equation 3.0 DS Equation Equation.39qP  2  FMicrosoft Equation 3.0 DS Equation Equation.39qCompObjfObjInfoEquation Native 2_1075619998F > >Ole CompObjfObjInfoEquation Native r( r(x)=sin(x)/xx"["10,10] FMicrosoft Equation 3.0 DS Equation Equation.39qf +ss\A_ŀ΀H9@3q=I\ GnOhL \L-@[4T[ n``Ĥ\Y\E肹L 2"{2wDdX  c &A?"`2Kg8=`ؾ'jԀ`!g8=`ؾjxcdd`` @c112BYL%bpu .| z- `+KRsA<.,BF0( J-fDdX  c &A?"`2Kg8=`ؾ'j`!g8=`ؾjxcdd`` @c112BYL%bpu_CH9@ڈ4HƴOM/ v" \?o qr5#}`ܝ äݹQT 0ba&R ?׀yA9M gi]Ă" &T`؀2$$d! =kQz|+;  W&0;2q. p{EQ;@|]8+73o-X5+RQf&JSl#$\ ! v 0#Ĥ\Y\ t0.?0jDd Vh  s *A ? ?3"`?2*F|X/a֑ΔſY`!F|X/a֑Δſ:@@xMN P =U 'ঈS?@~@UAA#$XspnN@цyEA?rF>5Zs'º:ZH7a]#EKA-8U"S'_zb:4:zmn0? nK$C4268%vb%:pQ <dfoAR,h .IDd hUh  s *AC? ?3"`?2\Q|429(_8?`!0Q|429(_`@:|xcdd``~ @c112BYL%bpu .| z- `+KRsA<.,BF0( J-fDd P  S A)? "`2,ֵQ6/]'y'`!ֵQ6/]'y4@Yxcdd`` @c112BYL%bpu .| z- `+KRsA<.,BF0( J-fDdX  c &A?"`2Kg8=`ؾ'j`!g8=`ؾjxcdd`` @c112BYL%bpu .| z- `+KRsA<.,BF0( J-fDd_X  c &A?"`2Kg8=`ؾ'j`!g8=`ؾjxcdd`` @c112BYL%bpu .| z- `+KRsA<.,BF0( J-fDdJ  c &A?2Kg8=`ؾ'j`!g8=`ؾjxcdd`` @c112BYL%bpu .| z- `+KRsA<.,BF0( J-fDdX  c &A?"`2Kg8=`ؾ'j`!g8=`ؾjxcdd`` @c112BYL%bpu .| z- `+KRsA<.,BF0( J-fDdX  c &A?"`2Kg8=`ؾ'j`!g8=`ؾjxcdd`` @c112BYL%bpu .| z- `+KRsA<.,BF0( J-fDdJ  c &A?2Kg8=`ؾ'jձ`!g8=`ؾjxcdd`` @c112BYL%bpu .| z- `+KRsA<.,BF0( J-f%Dd |Tbh  s *AF? ?3"`?2iɩ jpa)E`!=ɩ jpa)` 0XJ x]NAC/HaƂNcxk !K84\#4VWZR66p3?C' a#:".BZEl:'h0bsQF1kx&BFM?n@.yxJ=Uܣ?,yNn(糡w6D\Twԟi4dmd^Bat{fmqٛ/Fr6 V# =oDdJ  c &A?2Kg8=`ؾ'j`!g8=`ؾjxcdd`` @c112BYL%bpu!6l<m@g뭪-~j)^-6Nim&Uۗ?u}R?wE[~IOU~]# BQO՟=Uֈ7-0:zX JM''Q`+c>20E5((e(QNN V(((QPD@ye J)JqS(P+O!8ː^5H3(Ezһ#=HoHHHHHHHHHHHHHHHW-@K-*-\[ZOY]k~U=TgxֽtUVJ+;g̟<|UeaYWöO܉2Ro ^7K#nTdM?tuf~Fu55I5h>hHA{l뭽~^kkAj}zw㣫-N޼-ةMʷj"dh3F/7~~";C!~}⋦[0V?o=,jT.?ۧO}CN'}XT=w&jO3>m^fz}i!麔?zڬxT|=x|&ʷzlbu]Ytfy8Goe |&孒ʷkbv9U]B:o룿NY-q%6Kwt;zp:ojol_0Q?|v]qt_u!U]5/ma۬O>YyxvibY]EyY{|.xOwYLlg9q,\MPrə+E{51t<4&$:h21GSJCϴh21w-Do4I]Ф-LTдHM&hR M*?ޢMkQF04mF9֫&jdb]:j:&s4U 04|E9^udbw/COh21 M|M~ M*c&s4O245F9"CE9 CS;o4:/chRM&h F9B M*ȢM ?7LcT~E9W1w FYQ M*ҢM!c.CXo4) urG"|9\lCS7LѤb34M&^s4[M&hIIY4i!ϭWro4i1bhR M&hHp}5mF95L+Tz&s4G:h21G%Cʷ^Gab7>A&hx`o+}L__M??80G9R,LH 7LuФߴh21Gx*T`j uФY4cCSWo4ԱIoh21G_} MudbGQk Mjn&s4 t}5sss4=%y0TM&hzuLghF9"jR9M&hJA14MF9&T#M&hRIeh/әi &s4-r7Mktdb&um&E/6vM:X>UuΧPOcl+fQA.|k[gDD !OYEωho"/F "Q(+ú ̫VIsq[$>{PSB{b&UtLЇb!}, Nb,k޿;{h~jL)^["մtmnԞQG*;(: Lt/%R7K03QIN Dߨ=Ao`y`: F6p%b,(%b)48L`5BLDNAOQ"BQ"F`FbLD%EfelhBoGO^2hɻ'^> 4>(:AIM)N<^hя|4P߷Ƅ:^UF j*R Zt5X^vr"wmr;ur3unU.e=(S9z̢r2iL28zJFS a&GSE(12EIxJ)X?Q22e$eA2$''Nr$3iN˱U2c|=O_YӳssF>y0JcdLp8#c28hJdpa<8qЅ ca,8tC8HhJa28Lp8a8×2Up8_p;8܁Q;vp n+i58\ /?a*8La7~?ڷWp8Гѣhe0-0 FC8a:8 s|pXWÍzxVïa%z9ZzZv=ap810Ys8Jp 8 _a>pxCrpX2p&8Ýna18\ Åp8p8'tp8ca8ca88 pp8FHp ca8LIp8LTp8'Vp<rWV#z3S@\g55IN?lR?O+s[-L[505:8~ndƌdͻk?uӯ]yrMQ1>&|{N4zvOYע{GWo#0̑U űH.3saeYnf{ߗp_,duCM}^"K%=2B'GbR66foV~Ir_rtxk~ k5o#>8Z Q"VtHv-էWYDD4.\ڟø0Ŀ0:L_xfJYOcwz[s=|ךC{)ճz{m MMouR}CidbOgn e> .| z- `+KRsA<.,BF0( J-fDdJ  c &A?2Kg8=`ؾ'j`!g8=`ؾjxcdd`` @c112BYL%bpu .| z- `+KRsA<.,BF0( J-fDdJ  c &A?2Kg8=`ؾ'j`!g8=`ؾjxcdd`` @c112BYL%bpu .| z- `+KRsA<.,BF0( J-fDdJ  c &A?2Kg8=`ؾ'jM`!g8=`ؾjxcdd`` @c112BYL%bpu .| z- `+KRsA<.,BF0( J-fDdJ  c &A?2Kg8=`ؾ'j`!g8=`ؾjxcdd`` @c112BYL%bpu .| z- `+KRsA<.,BF0( J-fDdX  c &A?"`2Kg8=`ؾ'j`!g8=`ؾjxcdd`` @c112BYL%bpu .| z- `+KRsA<.,BF0( J-fDd_1089794714F#>#>Ole CompObjfObjInfoEquation Native _1090218501\F#>`>Ole CompObjf%e( (x i ,y i ),(i=1,...n) FMicrosoft Equation 3.0 DS Equation Equation.39q ²  "  2 =ObjInfoEquation Native _1074671072F`>`>Ole ,nn"d"1n(y i "2y  i ) 2i=1n " FMicrosoft Equation 3.0 DS Equation Equation.39q (t dCompObjfObjInfoEquation Native )_1074686041F`>`>Ole CompObjfObjInfoEquation Native > FMicrosoft Equation 3.0 DS Equation Equation.39q" d=nk FMicrosoft Equation 3.0 DS Equation Equation.39q_1090164640aF`>`>Ole CompObjfObjInfo ²  "  2 =nn"d"1n(y i "2y  i ) 2i=1n " FMicrosoft Equation 3.0 DS Equation Equation.39qEquation Native _1090164669F`>`>Ole CompObjfObjInfoEquation Native _1090164684F`>`>Ole  ¥ؚ  =nn"nk1n(y i "2y  i ) 2i=1n " FMicrosoft Equation 3.0 DS Equation Equation.39qCompObjfObjInfoEquation Native _1075618952F`>`> !&+059:;<=>?ABCDEFGHIJL ˜`P =kk"11n(y i "2y  i ) 2i=1n "  FMicrosoft Equation 3.0 DS Equation Equation.39qOle CompObjfObjInfoEquation Native < // s FMicrosoft Equation 3.0 DS Equation Equation.39q ¡ "  2 =1.51n(y i "2y  i ) 2i=1n "_1090164576F`>`>Ole  CompObj fObjInfo Equation Native  _1072617913 F`>`>Ole CompObje FMicrosoft Equation 3.0 Equation.2 Equation.39q `AJ e FMicrosoft Equation 3.0 Equation.2 Equation.39qObjInfoEquation Native <_1072617905F`>`>Ole CompObjeObjInfoEquation Native <_1089795358F`>`> `AJ e FMicrosoft Equation 3.0 DS Equation Equation.39q%! ["1,1]Ole CompObjfObjInfoEquation Native =_1089795465F`>`>Ole CompObj fObjInfo" FMicrosoft Equation 3.0 DS Equation Equation.39q% ̍  FMicrosoft Equation 3.0 Equation.2 Equation.39qEquation Native #)_1072617912F1>1>Ole $CompObj%eObjInfo'Equation Native (<_1072617909F1>1>Ole ) `AJ e FMicrosoft Equation 3.0 Equation.2 Equation.39q `AJ e FMicrosoft Equation 3.0 EquatCompObj*eObjInfo,Equation Native -<_1072617907F1>1>Ole .CompObj/eObjInfo1Equation Native 2<ion.2 Equation.39q `AJ e FMicrosoft Equation 3.0 Equation.2 Equation.39q `AJ e_1072617906F1>1>Ole 3CompObj4eObjInfo6Equation Native 7<1TableSummaryInformation(8DocumentSummaryInformation8@X  c &A?"`2Kg8=`ؾ'j3`!g8=`ؾjxcdd`` @c112BYL%bpu .| z- `+KRsA<.,BF0( J-fDdX  c &A?"`2Kg8=`ؾ'j`!g8=`ؾjxcdd`` @c112BYL%bpu .| z- `+KRsA<.,BF0( J-ftDd hRh  s *AI? ?3"`?28زQz *̏`| `!8زQz *̏`, @(|ZxڥR=KA}wIMJ"XaVZXkC^"$R KCl-ƭ#eMj*-R^rx?a8a ̨Y2ްxPV,ކӯ:&Mu1cus?>[ KFwDgδΈni܈PHuUؙ(SzqH~{Sc-)-bQ̈O:K7~Dd P eb  c $AJ? ?3"`?2?k]နbm`!k]နbm@xڥT;KA.HDt6>(؉b@!BB6 ZXV`" J,=uDx7;w{K @||бmm[ZݤYn\F!vZ[|$9x|[L@{KV#Awq [~Ƿp|+AQ+՛hw,2y'!h |ףϼ£g/n3|?E:KթxQp;""%[2'[gk:2\|8f3S P[%-07Z_ ouTq^'"oW؋-iS3.ޥ rf(\e?C"=z<d'*p'U˿SbQl oufqT|Fq[l9f'Zr ѱ&0noSŶ\АD.v0o8|+KRsA<.h0]7]LRFDd P eb  c $AM? ?3"`?2>u U9|n*`!u U9|n*@xڥT;KA=B"X" l|`' BQB:+VXG #zS ػfgno (CfWat-$D[ǑViԾga(D&pD0D9.rk81 N,$| B[FB\Dq7` רup}A׷?aYE# B>y>xȏAd\Cd^ Cb)߲eNWCxK?,"4o:/S-%":/5mCo H3\l(d3s0,cnRǴ7R_Gi`^?n_MϯC zԫG9#۵zx(\aKyu,^^Km^v_g@Wnh|?ҿӣ͒7L}:QT(%:gAlIB*#O ,_i@jDd eb  c $AN? ?3"`?24~KBʹI8`!~KBʹI8٦ x xڥTM(a~awv E@QpDMR{%\̅\Apsqۋ R.f _fyg}QF j_:[Q#$DE=Q QX3,/)$x Yf`QccBHƝvHeZ䥱Lvi[G,|n(\ 4?ѹ'<~opDu'!3q֓~ EUAcʸĆsJ=No4Q toP³< e1[jrFsA8ǩbǿ3frF࠘B0kGcZ}D* t(.NMG:WJDd \eb  c $AO? ?3"`?2gw'j}}Bثgz`!gw'j}}Bثgzj` @0xڥSKPw/mLC*"`biVJ (:7dݡ :(߯Ĩ?hKܻ{9q:=[q-$DY=R qm.ά6C2Q1LΗrΪp1}F_[)"sw6VҚ[|\W{Nf{q0ӉSe-"Ҋ~֑L|χ佧}-|Ϲ޼>}~?.Wh7*%o|HТ/MzFuZDDK{Ej\M oP5=NIRTyʣDwˣLqO@>+c)ygJKK˩f̘Af͢?>-\/^LTWWGb jllUVQSS]ڨ:::hӦMI[l.۷Ν;iݴg룽{t!:|0ѣGĉtIӧO 9sΞ=KΝ/ҥKtz*]v͛422B4(B `qtw%s(A҈&Օⓟ}O7Lj?"{|Ofʿ^)I~格o.9& nA<)?[t}ޖ$AlM3_+EQ_}D+űxs٤i|FiCFI=_v+xYi]i|9G3S9)w|<Ƀ}_;GW#rٿNHٻH}ע Rx X6Z>c ϿAzEs]JNcݮsi~os?{,:fW#Ұ,b8۸3l|'|Iݦ<۔ G2'0l٫{Cx=sETOWJ'ToB85z=Oۜw_kxB9ϳ'|dxnOlwpǃ'k&<Om<)upz{2'eB\]W[ݣSmf OHJWԵV & mOl9y] A#`ɺ~!un0Oai5kXBA.d]ۄ_ׄqniBQqW¯k&\獣IJ>e%KyĺLaDK\3Ohέs8p-ĵ(k\';Zs27(?\Dզ68I&ן*\Iyj/=vpv?;ĵSm \;uQ\75=?' ʹjP=k׮sRּqϛ/xs^pQSDkn5r<}x&QjA{{{s9tsQxjs7 =JDu>dMm6@Md#S=3sS/БR!F^E3|mx$羾x^x+.Ѿ)dMm6/]]]9y=1]1{w\l9i:l\챹N9=m 9gm3d#9,:k\0ló\_QSq+ J;H6?G tm~潚<='~~fߖ.Б-%V p=73|mx68^<ߨ<G;_q2nrθx|<o9xC7sɎ%~pbB{o9x[YQ05I!5?`<$~~fߖj БyݡeߖiB:n߾=e b.xtO4h_2Y>7|Q^q]#'f2 نgHbW*dc@Mvi-Fi{.N?/_}l Uu 7ϳ7>@ ݟu: s:ILM@G$ayU*yJ Бļ*Nb.xޭtO4h_ Ϧ6GsW=y3t*B:W= vL6.x6Ҿ(m/_;;;s*gX>qkb Ϧ!ް rγ?3iϟ`j?Q$a67fMk<x&qIU>JNNS/Б:Ɂrܤz}v(_;VM0ló)G4ȫUaγ?3iϟ`jɲNg6?Vq@^Z*y+-l3t$1 ޥL6$k< ig"nPy¶@ &>f&ic$d#s5a3֪k>%u~v~LszzIz 6WCm5t$^6~T4sgm!幥%'ϵ ^%DI\̿<'` نgSdM㮋ժb..S/БĺXXqNu3٘_Z*9ɯLk=x&qK|+P/a^/Hb$l~^rNe1f@d6ҾO~&q Tީ8(|F WyUM*;sqGi{.*O@yؔ}ۤa نgHbWP#rU3Ļ}{Wm! I̫3nU@&k̵ Mkz3Js񻌏oO xciJgS웾_ @.u؟~Oud]{'>\g@;9Ja3Zs>z|mx6MI>=q}{B Бq;jͤJL9yyϦ\6A{]ZXdӾ.S/Бuɰmy<$FcLv,sp[*o{۟ k^>)ϟ&}}̙3tY:w]x.]D/_+WիWڵk4<@=P5< %! 8 :@u!f0)0X@V0ȝr  W&0;0q-bF:`n2dVBSϼ\?},Ra 5wr'\& L\ 0~=DYd++&1_ `Us NPpTzNݿӿA**́_\ q'[7s17!|,K/UjO.;_gaEvgS5 o Kzp~43\Y  fG\50ɀ-t.piK.?+KRsA<.hg+̘m"YDd aB  S A)? 2,ֵQ6/]'y:`!ֵQ6/]'y4@Yxcdd`` @c112BYL%bpu .| z- `+KRsA<.,BF0( J-fDd-!NN^  C ,ASFig10_new"`"XH8Z<@=XH8ZtAoSx}pTs V#:F!T:8AN D:P@`[t  Pjb(l"Fk,J" PH =7y{dٝ9{/`b渱sp(ykWc$cH 0Q*@_u A|g<! =1c\ c=H;c lCA10r011*0 F,h0R1cda`ccT`Tc44`zWczcz>`z/TL1}4b`z/`:`:`:`:`:`:`:`:`:`:`:`:`:`:`+ TTlWcW+o=c0=Ga(7f@{)uk5`{\c)oO5NͫOyׇ Mѯ?f("g5:>>o85 ޫYoUl"~ؔ C |`=o}7-FOlTu~8Kg5N7DkV@nʟz=}2L$mn(X)<3>7~ԜKۡM>nI=CFXM8ڶ:h~~Ư011 [lCY͍?A{Ü鮿iAG .dcbLܸ=G1^gϱޯj˩sz-Wb ޻\]jn}v:';S6S@>Yg-S@>>ūS9url΅:1-gsu2,%c*gsud19mN>3Yg2S@>3YgS@>5c1)_g{lΏNG_g{lLfdLuϖLd֙HƔ(Pg{lNNORu:u1 >[s)ΥdLKlo-ՙ3)AR1>s/оs/8&uEuwl͓q)ڙ'KikWWCs1-} 5XKv[ X2kuuuE]j]_ZisIpu\?Zߩu_stkXi5BLzA9s`=ׁ֥֥LRb]*`[zy$6XFkkankͽ>DZ>r-k}%gҞKZc}W%:Yz`}w%ԑ\<^G!q7Zo= CXohf/ǹ?r&sb K}}T"kRR|is}9c%sLɬkW u^s)|-is3֙:Sz TǜK#6:iHBXWs.{p\kG}oYu̹2~is9c% k : ZW׸9½_8muӺX XvSz;t̹=S~is2֫_BXs.{ߘr\uNNkcRmu :XK~C :XX; M;u4iqkر eYJ.JX >_gJ 8AgL^@zfITpBz V+PZ eU\ԛWm۰Gm;P}U*-jrxK} 1x]}kT RaRjՋ^=65C UOq5RMR߫GԣjzXQRS٧=j^Rwt^U7C.{{PaJ0tʡ)@k m{mg0{u֐k=5c[qw]zßzƜNεLg2|ްyy|^ "Y=v>t>G 8t@/xΰsӹ8 8e:78ve:ZLz\9nGۑYϑ"?We3079ug='GcU19mUUĹJ|8I''ILOv>t>AO8t>IOzaLLwp>j((q>*ӹ8WFsd=_&SO1OSUL3k ;0kss5,q>*sep~.)?ܜwis1&טi%L\"༗\F<˰..K9\D<\`ع\@ w0 sv>t>H 8g:Isa2sq.p.a:"·<߰~~_y q>}1}e:'=\jع\JK?e:r8v.g:rRsq0\A07v3N=ƘiΘi罆23Ks7vtD7 8g3p^gyyq^'t^O{yaLy+|6=oq4V!u19 tdء:~A~lk:< <Say? ,0nq%FW6ݾn;_mows=/[wvZq`5G2]fJb\6ו.LGWknd_|/(j߈1Ç1c"4TY9 0+01=ӳ0}1b4L1=}>{.u/+J|MMwDdeJ  c &A?2Kg8=`ؾ'jON`!g8=`ؾjxcdd`` @c112BYL%bpuط3f,X6)HU(Gi*P3ܟ SC_Ԙ M>:I2Zϒ} 5ǷIev z9IR!lBGgb&N&;PRv݉$(fl}O}3HCDK,DdLX  c &A?"`2K@lt>@lt>՜.+,D՜.+,t0 hp  University of MinnesotaS_A 5Robust Parameter Choice in Support Vector Regression Title 8@ _PID_HLINKSA umailto:myq@ece.umn.edu<V mailto:cherkass@ece.umn.edu  FMicrosoft Word Document MSWordDocWord.Document.89q      !"#$%&'()*+@@@ ?NormalCJ_HaJmH sH tH J@J Heading 1$$dh@&a$ 5EH\R@R Heading 2$$@&]`a$5CJT@T Heading 3$$d@&5CJ \aJ :@: Heading 8$@&5\DA@D Default Paragraph FontViV  Table Normal :V 44 la (k(No List 0a@0 HTML Cite6]>B@> Body Text$7$8$H$a$4 @4 Footer  !.)@!. Page Number<"@< Caption$dha$5\<P@B< Body Text 2$a$5<Q@R< Body Text 3$a$CJLC@bL 3Body Text IndentxVD^jsj ~Z Table Grid7:V0VR@V s Body Text Indent 2dxVD^4U@4  Hyperlink >*ph  !"#$%&'()*+,-./0123456789:;<=>?@ABCDEFGHIJKLMNOPQRSTUVWXYZ[\]^_`abcdefghijklmnopqrstuvwxyz        !!""##$$%%&R`'a(b)c*d+e,f-g.h/i0j1k2l3m4n5o6p7q8r9s:t;u<v=w>x?yyNAOBPCQDRESFTGUHVIWJXKYLZM[N\O]P^Q_&'S(T)U*V+W,X-Y.Z/[0\1]2^3_4`5a6b7c8d9e:f;g<h=i>j?k@lAmBnCoDpEqFrGsHtIuJvKwLxM@z  !"#$%&'()*+,-./0123456789:;<=>?@ABCDEFGHIJKLMNOPQRSTUVWXYZ[\]^_`abcdefghijklmnopqrstuvwxyz}      !"#$%&'()*+,-./0123456789:;<=>?@ABCDEFGHIJKLMNOPQRSTUVWXYZ[\]^_`abcdefghijklmnopqrstuvwxy>QRv}~ - NO? h;rl[vK t_#$&J'+,^--M/N/0o00'2(2U2V22 37334z44a58d89+<<<=> ?@@?@@@@CCElFFGGGI(IIJJ[K\KdKKKGLL4MM_NNO!PPDQQgRhRiRjRlRRR7TUUUUUVVUZt[u[w[{[}[[[\\\\\\!\"\\\````aaa-b6e7e9e=e?eCeDeeeeeemfnffSggh iiijQjmmmBnCnn(ooWppqqqq0tvvwwwx7xx`yayy z}}}U~V~~02‚ʂUV܃8ׄ`*ևׇ\n9Pqrtɔ0@ACDژۘf\ƬǬȬɬʬլ֬gWYQdxy  !"#$%&'()*+,-./0123456789:;<=>?@ABCDEFGHIJKLMNOPQRSTUVWXYZ[\]^_`abcdefghijklmnopqrstuvwxyz{|}~00p000000000000000000@0@0@0@00000 0p0p000000000000p0p0p0p0p0000p0p0p000p0p0p 0p 0p 0p 0 0000p0(0p0p0p01p01p010101010101010101010101p01p01p01p01p01p000{?p0{?p0{?0{?p0{?0{?0{?0{?p0{?@0{?p0@0{?p0{?p0{?p0{?0{?p0{?0{?0{?0{?0{?{?0{?0{?0{?0{?0{?0{?0{?0{?0{?0{?0{?0{?p0{?p0p0{?P0{?p0{?p0P0{?p0P00P0{?P0{?p0P0P0{?p0{?p0{?X0{?p0X0p0X0{?p0{?p0{?p0{?X0{?p0X0p0{?p0{?p0{?p0{?p0`0p0{?p0{?p0{?p0{?p0{?0{?p0{?0{?`0{?p0`0p0{?p0{?p0{?p0{?p0p0{?`0{?p0{?p0{?p0{?p0`000{?0{?0{?0{?0{?0{?0{?p0{?0{?0{?0{?0h0h0p0{?0p0pp0pp0pp0pp0pp0pp0pp0pp0pp0p0p0p0p0pp0pp0p0pp0pp0p00x0x0x0pp0px0pp0pp0pp000pp0pp0p0p0p0p0p0p0p00p00p00000p0p0p0p00000000p0p0p0p0p0p0000p0p0p0p0p000p0p0p0p0p0p0p00p00p000000000000000000000p00p@p@p@p@0@p@p0ņpppppppppppppppppppppppppppppppppppppppppppppppppppppppppppppppppppppppppppppppppppppppppppppppppppppppppppppppppppppppppp0"\O900@pR  )S!#u%&d*-W0)3V9<>/@AB?FbIL0PVQSUVbXZ[]F`tcdghOkZlmnpVrs$u;w:y}hl ]ǒ%И=fKg R befhijkmnopqrtuvwxz{|}~ &V:EO!X^hmnw΀׏d5Rocglsyd>OQ(=Pvx ? P R Ufhq ">OQ`qsEVX 13;LN7HJ,=?art}K\^hy{}    |!!!###$$$%%%'%q%%%%%%@&S&U&&&&')'+'['n'p'C(V(X((((();)=)P)c)e)))))))K+^+`+/,B,D,.+.-.000B1U1W123 3 3335373J3L34(4*4444444566N6a6c6666666888j8}888889999999::,:?:A:|::::::n;;;4<G<I<<<<===)><>>>>>>:?M?O?@@@OAbAdAAAA6CICKCuCCClFFFGGGGHHH0H2HHHHCIVIXIJJJJJJKKKKLLMLaLcLLLL~MMMN"N$NNNN>OQOSOOOOcPvPxPPQQQQQR2R4RoTTTT U U U3U5UUUUVVVVVV3XFXHXYYYZZZ\]]{]]]8^K^M^^ _ _v______#`6`8````aaabbbbbbcccddd%d8d:dFdYd[dgdzd|deeeefff2f4ffffSgfghggggh2h4hhhhhhhLi_iaij!j#jWjjjljjjj kkk(k;k=kl,l.l:lMlOlZlmlolzlllmmmm nnlnnn(o;o=odowoyoop pWpjplpppp$q7q9qrrrtttiv|v~vvvv$w7w9wSwfwhwwww7xJxLxxxxyyy{.{0{{{{ |3|5|V|i|k|} ~ ~&~9~;~V~i~k~\oq#68DWYatv.08KMt`su+>@džچ܆Vik J]_ۈ\oqNJڊ܊n9LNZmo’ĒɔܔޔSfh+<>9JLWhjZkmХ˯ޯ ::::XX::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::: !!T@/Xb$SJq.MS(R$5e ogZ>()1@r( n*urn:schemas-microsoft-com:office:smarttags PostalCodeB*urn:schemas-microsoft-com:office:smarttagscountry-region=*urn:schemas-microsoft-com:office:smarttags PlaceName= *urn:schemas-microsoft-com:office:smarttags PlaceType ! !!!! !! ! !!! ! jqo w " %  4*0vycf(())))++1188_8c8d8i8j888888H3H5H6H>H?HCHKKKKKKLLRRUUUUaa(b,b-b4beerrrrss zzjl\rƊ9OZpɔߔ+/02t|"ݬX[mqs{#(22:?ENTVY kpu{X]mv =Fų$).7=ƴlr<DQZ >ARTN T jq?D 447wz!,[^vz&)fhx!|!""""[%a%%%''++-- . .z4|4556688c8d8j8899999::<<oCuC~EEFFGH3HH!I&IKKKFLGLMLlRsRUUVVVTW0X3XxYY[[[["\)\````aaaa,b-b4bDeKeQjVjkkttyu{u{{29V\DJ\rƊHQ9OZpɔߔ+DKĢƢǥХDG<ٳMO϶׶ 33333333333333333333333333333333333333333333333333333333333333333333333333333 488c8j88GH3HHKKKFLNPRjRlRtRUUw[{[\\``aa,b-b9e=eeegioq02:ׇ\r9OPTZpɔߔ+DLSt 488j88H3H5HKKaa\r9OZpɔߔ cherkassky cherkassky cherkassky cherkassky cherkassky cherkassky cherkassky Unknown User Unknown User mayunqian O s]>%}x!Ii%I=`4rUMGFSRn9[ $N`i\ $N`0j4}hh^h`o(()^`o(()  ^ `. L ^ `L.hh^h`.88^8`.L^`L.^`.^`.xLx^x`L.hh^h`o(() H\H^H`\hH) \^`\hH. \^`\hH. 4\4^4`\hH)  \ ^ `\hH. | \| ^| `\hH.  \ ^ `\hH) \^`\hH.h ^`OJQJo(h ^`OJQJo(oh pp^p`OJQJo(h @ @ ^@ `OJQJo(h ^`OJQJo(oh ^`OJQJo(h ^`OJQJo(h ^`OJQJo(oh PP^P`OJQJo( \^`\hH. H\H^H`\hH) \^`\hH. \^`\hH. 4\4^4`\hH)  \ ^ `\hH. | \| ^| `\hH.  \ ^ `\hH) \^`\hH. H\H^H`\hH. \^`\hH) \^`\hH. 4\4^4`\hH.  \ ^ `\hH) | \| ^| `\hH.  \ ^ `\hH. \^`\hH) h\h^h`\hH.^`o(- hh^h`OJQJo( H\H^H`\hH) \^`\hH. \^`\hH. 4\4^4`\hH)  \ ^ `\hH. | \| ^| `\hH.  \ ^ `\hH) \^`\hH. \^`\hH. H\H^H`\hH) \^`\hH. \^`\hH. 4\4^4`\hH)  \ ^ `\hH. | \| ^| `\hH.  \ ^ `\hH) \^`\hH.\^`\CJOJQJaJo(hHlH\H^H`\OJQJo(hHl\^`\OJQJo(hHu\^`\OJQJo(hHl4\4^4`\OJQJo(hHn \ ^ `\OJQJo(hHu| \| ^| `\OJQJo(hHl \ ^ `\OJQJo(hHn\^`\OJQJo(hHu s]Ii%O0jI=i\n9[GFSrUM}x! @                   N!I!Q!r"""%"x"-#9#V${$<% &3&+'l'1p''(,(-4(+*.*Q*]*gx*+.;+r+#t+ht+ ,9,0C,U,s-7.?a./.$/ I/1\1X`1@2_t2v2_34 4F4Nq4 5 +5 6K6}6B 787497<7@7r8_89'9i59989U:A:W:#;Z`; <<+<F-<4<bR<X<j<}<=:=R=k=t>>R> ?:?a? @A A&&AI|A+B.B\ByC CaCaCD)$D\HDzD-EHTE[}FFzGG HH+'HnH?_aje AWyd@3Nu~-26D71m=9o]vy-G0B EQSW^0@d4f_ /v_0<Sz"y#GO&UoOHTu\x5/3KPSjt|8o9RjoHSIe w l/@?%.AD+5%e&s*6)r.~Y`{`3r3sH"x-oKf&39G\xc#Qy  :h2o"#O*`]7QU)VKmed8Sg@d P@UnknownGz Times New Roman5Symbol3& z Arial;SimSun[SO?5 z Courier New;Wingdings 1hBhBhBh_S_S!-!),.:;?]}    & 6"0000 0 0 00000 =@\]^([{  0 0 00000;[d 3QH? c4Robust Parameter Choice in Support Vector Regression mayunqian mayunqian4         CompObjKj