%PDF-1.3 1 0 obj << /Kids [ 4 0 R 5 0 R 6 0 R 7 0 R 8 0 R 9 0 R 10 0 R 11 0 R 12 0 R 13 0 R 14 0 R 15 0 R ] /Type /Pages /Count 12 >> endobj 2 0 obj << /Subject (Neural Information Processing Systems http\072\057\057nips\056cc\057) /Publisher (Curran Associates\054 Inc\056) /Language (en\055US) /Created (2018) /EventType (Poster) /Description-Abstract (Time series classification using deep neural networks\054 such as convolutional neural networks \050CNN\051\054 operate on the spectral decomposition of the time series computed using a preprocessing step\056 This step can include a large number of hyperparameters\054 such as window length\054 filter widths\054 and filter shapes\054 each with a range of possible values that must be chosen using time and data intensive cross\055validation procedures\056 We propose the wavelet deconvolution \050WD\051 layer as an efficient alternative to this preprocessing step that eliminates a significant number of hyperparameters\056 The WD layer uses wavelet functions with adjustable scale parameters to learn the spectral decomposition directly from the signal\056 Using backpropagation\054 we show the scale parameters can be optimized with gradient descent\056 Furthermore\054 the WD layer adds interpretability to the learned time series classifier by exploiting the properties of the wavelet transform\056 In our experiments\054 we show that the WD layer can automatically extract the frequency content used to generate a dataset\056 The WD layer combined with a CNN applied to the phone recognition task on the TIMIT database achieves a phone error rate of 18\0561\134\045\054 a relative improvement of 4\134\045 over the baseline CNN\056 Experiments on a dataset where engineered features are not available showed WD\053CNN is the best performing method\056 Our results show that the WD layer can improve neural network based time series classifiers both in accuracy and interpretability by learning directly from the input signal\056) /Producer (PyPDF2) /Title (Learning filter widths of spectral decompositions with wavelets) /Date (2018) /ModDate (D\07220190218225454\05508\04700\047) /Published (2018) /Type (Conference Proceedings) /firstpage (4601) /Book (Advances in Neural Information Processing Systems 31) /Description (Paper accepted and presented at the Neural Information Processing Systems Conference \050http\072\057\057nips\056cc\057\051) /Editors (S\056 Bengio and H\056 Wallach and H\056 Larochelle and K\056 Grauman and N\056 Cesa\055Bianchi and R\056 Garnett) /Author (Haidar Khan\054 Bulent Yener) /lastpage (4612) >> endobj 3 0 obj << /Type /Catalog /Pages 1 0 R >> endobj 4 0 obj << /Contents 16 0 R /Parent 1 0 R /Resources 17 0 R /MediaBox [ 0 0 612 792 ] /Annots [ 45 0 R 46 0 R 47 0 R 48 0 R 49 0 R 50 0 R 51 0 R 52 0 R 53 0 R ] /Type /Page >> endobj 5 0 obj << /Contents 54 0 R /Parent 1 0 R /Resources 55 0 R /Group 69 0 R /MediaBox [ 0 0 612 792 ] /Annots [ 107 0 R 108 0 R 109 0 R 110 0 R 111 0 R 112 0 R 113 0 R 114 0 R ] /Type /Page >> endobj 6 0 obj << /Contents 115 0 R /Parent 1 0 R /Resources 116 0 R /MediaBox [ 0 0 612 792 ] /Annots [ 117 0 R 118 0 R 119 0 R 120 0 R ] /Type /Page >> endobj 7 0 obj << /Contents 121 0 R /Parent 1 0 R /Type /Page /Resources 122 0 R /MediaBox [ 0 0 612 792 ] >> endobj 8 0 obj << /Contents 123 0 R /Parent 1 0 R /Resources 124 0 R /MediaBox [ 0 0 612 792 ] /Annots [ 129 0 R 130 0 R 131 0 R 132 0 R 133 0 R 134 0 R 135 0 R 136 0 R 137 0 R 138 0 R 139 0 R 140 0 R 141 0 R 142 0 R 143 0 R 144 0 R ] /Type /Page >> endobj 9 0 obj << /Contents 145 0 R /Parent 1 0 R /Resources 146 0 R /Group 177 0 R /MediaBox [ 0 0 612 792 ] /Annots [ 178 0 R 179 0 R 180 0 R 181 0 R 182 0 R 183 0 R 184 0 R 185 0 R 186 0 R 187 0 R ] /Type /Page >> endobj 10 0 obj << /Contents 188 0 R /Parent 1 0 R /Resources 189 0 R /Group 194 0 R /MediaBox [ 0 0 612 792 ] /Annots [ 196 0 R 197 0 R 198 0 R 199 0 R 200 0 R 201 0 R 202 0 R 203 0 R 204 0 R 205 0 R 206 0 R 207 0 R 208 0 R ] /Type /Page >> endobj 11 0 obj << /Contents 209 0 R /Parent 1 0 R /Resources 210 0 R /Group 177 0 R /MediaBox [ 0 0 612 792 ] /Annots [ 213 0 R 214 0 R 215 0 R 216 0 R 217 0 R 218 0 R 219 0 R 220 0 R 221 0 R 222 0 R 223 0 R ] /Type /Page >> endobj 12 0 obj << /Contents 224 0 R /Parent 1 0 R /Resources 225 0 R /MediaBox [ 0 0 612 792 ] /Annots [ 226 0 R 227 0 R 228 0 R 229 0 R 230 0 R 231 0 R 232 0 R ] /Type /Page >> endobj 13 0 obj << /Contents 233 0 R /Parent 1 0 R /Type /Page /Resources 234 0 R /MediaBox [ 0 0 612 792 ] >> endobj 14 0 obj << /Contents 239 0 R /Parent 1 0 R /Type /Page /Resources 240 0 R /MediaBox [ 0 0 612 792 ] >> endobj 15 0 obj << /Contents 241 0 R /Parent 1 0 R /Type /Page /Resources 242 0 R /MediaBox [ 0 0 612 792 ] >> endobj 16 0 obj << /Length 3037 /Filter /FlateDecode >> stream xrܸVq* ߷QfJl0$FØC|}I$rH@ }R F& 2?S8^!0̓MYfo~|Zvd>48ܖޯt[{Mmotx@׃xv_áqoЙg-jQ _6*I467*sfcV#ի۫/HI~lJXe9pzOhG:ԛw3D'oIn 8bo=|Fw~+3LҞL7m3,ex'i+*X~缉=aqؿ`CSxC5vA(knZ6k&|VY,dgܝ*{<["s?& $LaՓ$* puY<,FUٿ(JbM`wYDWfʟv=R1\@Yl<%IFۮh~Q 0h1l0,=#[2{F}y~,hz]j@Gܞ32zG>0g ^}u͐d;o C8,Ύ I4Q$?eӺp4%ЁD84 rS?=jDyQaTMQ]lWN;f
V^@3ha0Y^Bz͋y`/~(⫓7@o=rAb
5-|py0m]+ؑ^5#!w\ੳl͎V~̎3*kFP!Xe/Fp{X؏M1Jک6К_U2k *rT_<'it;K c1:x#'xojNz.cOUL5܁0e <%G