%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 ] /Type /Pages /Count 9 >> 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 (2014) /EventType (Poster) /Description-Abstract (In this paper\054 we develop a family of algorithms for optimizing superposition\055structured\216 or \215dirty\216 statistical estimators for high\055dimensional problems involving the minimization of the sum of a smooth loss function with a hybrid regularization\056 Most of the current approaches are first\055order methods\054 including proximal gradient or Alternating Direction Method of Multipliers \050ADMM\051\056 We propose a new family of second\055order methods where we approximate the loss function using quadratic approximation\056 The superposition structured regularizer then leads to a subproblem that can be efficiently solved by alternating minimization\056 We propose a general active subspace selection approach to speed up the solver by utilizing the low\055dimensional structure given by the regularizers\054 and provide convergence guarantees for our algorithm\056 Empirically\054 we show that our approach is more than 10 times faster than state\055of\055the\055art first\055order approaches for the latent variable graphical model selection problems and multi\055task learning problems when there is more than one regularizer\056 For these problems\054 our approach appears to be the first algorithm that can extend active subspace ideas to multiple regularizers\056\042) /Producer (PyPDF2) /Title (QUIC \046 DIRTY\072 A Quadratic Approximation Approach for Dirty Statistical Models) /Date (2014) /ModDate (D\07220141202160257\05508\04700\047) /Published (2014) /Type (Conference Proceedings) /firstpage (2006) /Book (Advances in Neural Information Processing Systems 27) /Description (Paper accepted and presented at the Neural Information Processing Systems Conference \050http\072\057\057nips\056cc\057\051) /Editors (Z\056 Ghahramani and M\056 Welling and C\056 Cortes and N\056D\056 Lawrence and K\056Q\056 Weinberger) /Author (Cho\055Jui Hsieh\054 Inderjit S\056 Dhillon\054 Pradeep K\056 Ravikumar\054 Stephen Becker\054 Peder A\056 Olsen) /lastpage (2014) >> endobj 3 0 obj << /Type /Catalog /Pages 1 0 R >> endobj 4 0 obj << /Contents 13 0 R /Parent 1 0 R /Type /Page /Resources 14 0 R /MediaBox [ 0 0 612 792 ] >> endobj 5 0 obj << /Contents 56 0 R /Parent 1 0 R /Type /Page /Resources 57 0 R /MediaBox [ 0 0 612 792 ] >> endobj 6 0 obj << /Contents 86 0 R /Parent 1 0 R /Type /Page /Resources 87 0 R /MediaBox [ 0 0 612 792 ] >> endobj 7 0 obj << /Contents 92 0 R /Parent 1 0 R /Type /Page /Resources 93 0 R /MediaBox [ 0 0 612 792 ] >> endobj 8 0 obj << /Contents 102 0 R /Parent 1 0 R /Type /Page /Resources 103 0 R /MediaBox [ 0 0 612 792 ] >> endobj 9 0 obj << /Contents 104 0 R /Parent 1 0 R /Type /Page /Resources 105 0 R /MediaBox [ 0 0 612 792 ] >> endobj 10 0 obj << /Contents 106 0 R /Parent 1 0 R /Type /Page /Resources 107 0 R /MediaBox [ 0 0 612 792 ] >> endobj 11 0 obj << /Contents 116 0 R /Parent 1 0 R /Type /Page /Resources 117 0 R /MediaBox [ 0 0 612 792 ] >> endobj 12 0 obj << /Contents 137 0 R /Parent 1 0 R /Type /Page /Resources 138 0 R /MediaBox [ 0 0 612 792 ] >> endobj 13 0 obj << /Length 3244 /Filter /FlateDecode >> stream xY[~_,E8F$m`+ͮK!q}sEj6R83g2VwhU便_)|ZXee^mWlòHWY^|_MXv,^eYFZ]Vo_<H*-dq9p=EP%n\ _?ri\*"ZoCxsH)QP2}t֪?#vچp6]6 "K5{ArKm'(=YUde3K课TVa7/VQXBvduFI*23TWiB}yVi0B6'MEfgQ:6槮sWq~?n֙, PQ:K`:[{ٴ^Q`hRz:@vEHbY84o^*/ *aИBM hw,II-ܠ2
a P#cx3e`|=mVFC=#=;_Ǜɴp}Hp:G\%O2<ӹԑz>`;t\E&*T4:6cG\%1=;A@Jiw 5_.ecڀh4m 'ɘC$@^>jj(~%*%``S4
63Yv f6Qqt+;N\q"~Xĝ!`"'3qęX"82[YdjV9gag[7FkdS[2/m[xǡݸjfhN
Y}dܙ! , ɞDx/%<_dF!LdZNnkh#lG sLߡ>bEZ< FnI3Gp%j'G֔P#B0Eh]Ks6cځ,>oxcR]2:fK9>[ =$#4[dIݭ@L\Lg@ϊ
^/$%'g$(p ~
n%+{!Y,őq7;wB榯{w!J2l*wN(qMX3,JqhiEE1y_A(N-mHi 3*h:mMA ;$wxζFM#=P68n@Zސe$T`7D,|epTDe\dZ
<V0ň^ݰE
zz9mǑ/X0OۚV67!qہU15~knh/ؔ0yF'\Vs˂jc10zhd|o
ھbAةCW2:JT0tʼnzb8y!ʱimEGAT/v c;;Y`4] eJOw Jڹeu:GYؐؓcBjx/+cVKal#ި싰445ySC-5ݢ=Ml\b4H"*g?#*d:-"l#gBJnb3R}*R*0/8j'~*q[d&`C"3"{y)y>q00v-[m>ϒ\u lSHiwFy)cs2KJ,?!pY
O;Y{70Jāi3o