From ceedb54d3b1392e15fe2514a5e168e0b6c3c121e Mon Sep 17 00:00:00 2001 From: kenneally15 Date: Wed, 6 Jun 2018 09:13:35 -0400 Subject: [PATCH 1/4] Output --- .DS_Store | Bin 0 -> 6148 bytes improved_wgan_training | 1 + 2 files changed, 1 insertion(+) create mode 100644 .DS_Store create mode 160000 improved_wgan_training diff --git a/.DS_Store b/.DS_Store new file mode 100644 index 0000000000000000000000000000000000000000..16e35f1b5972704f55082dd788d6f6401425e85d GIT binary patch literal 6148 zcmeHKPfrs;6n|4t%RgD5{E2wj(2Is1FpBi%QZNRLp|J#L0PAi$l!?ns-Cd|in)GJk z7w|5kegQAuG-RM7EmZb0jf>Wkt1RtV&H#cKV@{B8rjw>9@lZVC>>If>?-^cctwVbM7(Y1a!6_@$J*w|+hYiT`m_FTNJ zy}c8EMo;HzZe3J!fgNo68QWV{JyxQQwzGDy-mvF|aDUmhv%qMzU?ipl4{XM%cPMpb zk@1qKX2d2h`761w-Xbdn>jk4f>(MPt&DxwUFrPWhWx;k^TO5A{V{m6DolcqK>2xh+ z?o3RKrOazruh(kElX!gi^7!PPe5LmE`RK8+3dh{mz_XT>ee z8F|~_?2BO>u-4Jpb-vr^>Fw)D^z;w(4<`DDdM^wYj6TOMJ}7%~y;H&!UdG*|1*6+X zZfA;u>m@G*^ihywk8!B?w}iX2bnn15zg@wY#DPNdyQ37$VRnqN_~2XteGuL+z{l0H|FVg3U3e8i^wrM&HVHZRl6_Nr?;)ScrQ5d>mm ziQy{bKUWjvDt3|aa`MKvkm+!j?Z(!DaOo(LrLqlj)}u4OhOJ; z;33rE8SKIvcn6>0D}0B2(nSWy2pJ_QGEOFlMP|q>nIl<}53^&9r+!ga(?I0zIo4C( z%6Z%syo`B`nnz2Xq^B#zJ5G`uH?n<^ZL6=Kvz z3`7hZWuPVO8uI?X`S Date: Wed, 6 Jun 2018 09:20:04 -0400 Subject: [PATCH 2/4] Update README.md --- README.md | 33 +++++++++++++++++++++++++++++++++ 1 file changed, 33 insertions(+) diff --git a/README.md b/README.md index 66449fdf..c3d8d548 100644 --- a/README.md +++ b/README.md @@ -24,3 +24,36 @@ download URL is in the file. - `python gan_64x64.py`: 64x64 architectures (this code trains on ImageNet instead of LSUN bedrooms in the paper) - `python gan_language.py`: Character-level language model - `python gan_cifar.py`: CIFAR-10 + +# Improved Metric + +Algorithm 1 Battacharyya-GAN, our proposed algorithm. +Require: alpha, the learning rate. c, the clipping parameter. m, the batch size. n, the number of iterations of the critic per generator iteration. +Require: w, the initial critic parameter. theta, the initial generator’s parameters. + 1: while theta has not converged do + 2: for t = 0, …,n do + 3: Sample a batch from the real data + 4: Sample a batch of prior samples + 5: Calculate Battacharrya Coefficient + 6: if BC <= 0.5 do + 7: Wasserstein Gradient + 8: else do + 9: KL-Divergence Gradient + 10: end if + 11: Update w with gradient descent + 12: Clip w to be Lipschitz + 13: end for + 14: Sample a batch from the real data + 15: Sample a batch of prior samples + 16: Calculate Battacharrya Coefficient + 17: if BC <= 0.5 do + 18: Wasserstein Generator Gradient + 19: else do + 20: KL-Divergence Generator Gradient + 21: Update theta with gradient descent + 22: end while + + + + + From faa6527a1a013d38f717fbe7b3c550d9efc3122a Mon Sep 17 00:00:00 2001 From: Kevin Kenneally Date: Wed, 6 Jun 2018 09:21:14 -0400 Subject: [PATCH 3/4] Update README.md --- README.md | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) diff --git a/README.md b/README.md index c3d8d548..a82a87e6 100644 --- a/README.md +++ b/README.md @@ -25,7 +25,7 @@ download URL is in the file. - `python gan_language.py`: Character-level language model - `python gan_cifar.py`: CIFAR-10 -# Improved Metric +## Improved Metric Algorithm 1 Battacharyya-GAN, our proposed algorithm. Require: alpha, the learning rate. c, the clipping parameter. m, the batch size. n, the number of iterations of the critic per generator iteration. From ff96e2bf56f7880c1cff493a08d85fe632472036 Mon Sep 17 00:00:00 2001 From: Kevin Kenneally Date: Wed, 6 Jun 2018 09:30:20 -0400 Subject: [PATCH 4/4] Update README.md --- README.md | 27 --------------------------- 1 file changed, 27 deletions(-) diff --git a/README.md b/README.md index a82a87e6..39b1fcbd 100644 --- a/README.md +++ b/README.md @@ -25,33 +25,6 @@ download URL is in the file. - `python gan_language.py`: Character-level language model - `python gan_cifar.py`: CIFAR-10 -## Improved Metric - -Algorithm 1 Battacharyya-GAN, our proposed algorithm. -Require: alpha, the learning rate. c, the clipping parameter. m, the batch size. n, the number of iterations of the critic per generator iteration. -Require: w, the initial critic parameter. theta, the initial generator’s parameters. - 1: while theta has not converged do - 2: for t = 0, …,n do - 3: Sample a batch from the real data - 4: Sample a batch of prior samples - 5: Calculate Battacharrya Coefficient - 6: if BC <= 0.5 do - 7: Wasserstein Gradient - 8: else do - 9: KL-Divergence Gradient - 10: end if - 11: Update w with gradient descent - 12: Clip w to be Lipschitz - 13: end for - 14: Sample a batch from the real data - 15: Sample a batch of prior samples - 16: Calculate Battacharrya Coefficient - 17: if BC <= 0.5 do - 18: Wasserstein Generator Gradient - 19: else do - 20: KL-Divergence Generator Gradient - 21: Update theta with gradient descent - 22: end while