%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 ]
/Type /Pages
/Count 11
>>
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 (Generative Adversarial Networks \050GANs\051 are one of the most practical methods for learning data distributions\056 A popular GAN formulation is based on the use of Wasserstein distance as a metric between probability distributions\056 Unfortunately\054 minimizing the Wasserstein distance between the data distribution and the generative model distribution is a computationally challenging problem as its objective is non\055convex\054 non\055smooth\054 and even hard to compute\056 In this work\054 we show that obtaining gradient information of the smoothed Wasserstein GAN formulation\054 which is based on regularized Optimal Transport \050OT\051\054 is computationally effortless and hence one can apply first order optimization methods to minimize this objective\056 Consequently\054 we establish theoretical convergence guarantee to stationarity for a proposed class of GAN optimization algorithms\056 Unlike the original non\055smooth formulation\054 our algorithm only requires solving the discriminator to approximate optimality\056 We apply our method to learning MNIST digits as well as CIFAR\05510 images\056 Our experiments show that our method is computationally efficient and generates images comparable to the state of the art algorithms given the same architecture and computational power\056)
/Producer (PyPDF2)
/Title (On the Convergence and Robustness of Training GANs with Regularized Optimal Transport)
/Date (2018)
/ModDate (D\07220190219001721\05508\04700\047)
/Published (2018)
/Type (Conference Proceedings)
/firstpage (7091)
/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 (Maziar Sanjabi\054 Jimmy Ba\054 Meisam Razaviyayn\054 Jason D\056 Lee)
/lastpage (7101)
>>
endobj
3 0 obj
<<
/Type /Catalog
/Pages 1 0 R
>>
endobj
4 0 obj
<<
/Parent 1 0 R
/Contents 15 0 R
/Type /Page
/Resources 16 0 R
/MediaBox [ 0 0 612 792 ]
>>
endobj
5 0 obj
<<
/Parent 1 0 R
/Contents 47 0 R
/Type /Page
/Resources 48 0 R
/MediaBox [ 0 0 612 792 ]
>>
endobj
6 0 obj
<<
/Parent 1 0 R
/Contents 55 0 R
/Type /Page
/Resources 56 0 R
/MediaBox [ 0 0 612 792 ]
>>
endobj
7 0 obj
<<
/Parent 1 0 R
/Contents 181 0 R
/Type /Page
/Resources 182 0 R
/MediaBox [ 0 0 612 792 ]
>>
endobj
8 0 obj
<<
/Parent 1 0 R
/Contents 202 0 R
/Type /Page
/Resources 203 0 R
/MediaBox [ 0 0 612 792 ]
>>
endobj
9 0 obj
<<
/Parent 1 0 R
/Contents 229 0 R
/Type /Page
/Resources 230 0 R
/MediaBox [ 0 0 612 792 ]
>>
endobj
10 0 obj
<<
/Parent 1 0 R
/Contents 251 0 R
/Type /Page
/Resources 252 0 R
/MediaBox [ 0 0 612 792 ]
>>
endobj
11 0 obj
<<
/Parent 1 0 R
/Contents 253 0 R
/Type /Page
/Resources 254 0 R
/MediaBox [ 0 0 612 792 ]
>>
endobj
12 0 obj
<<
/Parent 1 0 R
/Contents 267 0 R
/Type /Page
/Resources 268 0 R
/MediaBox [ 0 0 612 792 ]
>>
endobj
13 0 obj
<<
/Parent 1 0 R
/Contents 275 0 R
/Type /Page
/Resources 276 0 R
/MediaBox [ 0 0 612 792 ]
>>
endobj
14 0 obj
<<
/Parent 1 0 R
/Contents 277 0 R
/Type /Page
/Resources 278 0 R
/MediaBox [ 0 0 612 792 ]
>>
endobj
15 0 obj
<<
/Length 4150
/Filter /FlateDecode
>>
stream
xZɒ+pDx`MmdDBQL78
2##WY}4AFLur-|?~qiMij᧗aءg1?~~dnvfngoسP3z#_H){WֻaRQ,Y|{eط]]n4ꪾo^yQFĹ?>{=Pus?Mܛ7:Gʒ{tE<7?xun:O}AFX'AtC7 B;B7 ߗ~U|
S?a
3
L |hRc qVgs(wBn?/ՇsSWs=-yqC@J(WqZ!绾ݹ徿o>77;Hk46zYN!kZ'}k'
O k0sݝa13xQkx|kN]2&O7g8+ExUVmq2~[|gyf{;_oviXYڤ3ޅU,pՀ{4 ͷ0a5VJRsZ^~;007{aAn16# =wiWo;FKZOE* Gt;B7r@qrݶ]S`N!m=OfAIbG l'qdُe]6EW]YƘe(lOoAmaŦsu=\9ޡxEwFÞbkatnA,j,Tv̱)55 V6^Q¾ƺ]]u[B4W~f&]=7#8!3]9ovrB+E*N1: $gA1-HԎ
d MB_Ŭm veDAd#%/%l)ab-$^Ks=jg⛇տ`KEW?_Y %qAhS
C7'5HA&&3%'tL6`Bfdː&6n~AP%PLeX~3/R, Ir9ÈZvvӥhx1BP
'>њbLX7toCH\<7V0!=5b_u;:4X NAct)<{I%X\h&Gj|`n -Q!W*e$MoB|
nK5D(y7
";o$yι[zU%l0eIu4i*M({C&
]dvİqh
0
ک B<\K/7LjAqmP ]z[t&uPԶ-?@(PVJcdh}!8ѐp6``9bM {20Cngc]|QūuǨT
rpYYDL67MŅbE`˜ 3= a2mڗ2wG8=hֱ'Icxw_cX :r2/