Genera&ve Adversarial Networks NTT - PowerPoint PPT Presentation

�� Genera&ve Adversarial Networks �� NTT �� 2020 � 6 � 10 � @JSAI2020 KS-02 �� AI – MIRU2020 ��

�� 2014. 4 � NTT �� • 2017. 4 � 2020.3 �� 2�� • �� CF GAN DTLC- GAN GAN -PF GAN -PF for STFT GAN -VC [CVPR 2017] [CVPR 2018] [ICASSP 2017] [Interspeech 2017] [Interspeech 2017] �S�� CycleGAN-VC r GAN † CP- GAN † NR- GAN † Cycle GAN -VC Cycle GAN -VC2 Star GAN -VC2 [CVPR 2019] [BMVC 2019] [CVPR 2020] [EUSIPCO 2018] [ICASSP 2019] [Interspeech 2019] �I��

��S 2014. 4 � NTT �� • 2017. 4 � 2020.3 �� 3�� • �� CF GAN DTLC- GAN GAN -PF GAN -PF for STFT GAN -VC [CVPR 2017] [CVPR 2018] [ICASSP 2017] [Interspeech 2017] [Interspeech 2017] �� I��N�� CycleGAN-VC r GAN † CP- GAN † NR- GAN † Cycle GAN -VC Cycle GAN -VC2 Star GAN -VC2 [CVPR 2019] [BMVC 2019] [CVPR 2020] [EUSIPCO 2018] [ICASSP 2019] [Interspeech 2019] ��G��N�� A�� 3��

GAN �� 4 �� → Q1. �� • – �� = ��4�� T. Karras et al., “Analyzing and Improving the Image Quality of StyleGAN,” arXiv 2019. �

GAN �� → Q1. 5�� • 2�� – �� = 5�� T. Karras et al., “Analyzing and Improving the Image Quality of StyleGAN,” arXiv 2019. �

GAN �� Q2. ��6��?�� • �� by Monet �� JY. Zhu et al., “Unpaired Image-to-Image Translation using Cycle-Consistent Adversarial Networks,” ICCV 2017. �

GAN �� Q2. �7�� • �� by Monet �� JY. Zhu et al., “Unpaired Image-to-Image Translation using Cycle-Consistent Adversarial Networks,” ICCV 2017. �

GAN �� Q3. �� • �� T. Kaneko et al., “Generative Attribute Controller with Conditional Filtered Generative Adversarial Networks,” CVPR 2017. �

GAN �� Q3. �� • �� 9� http://www.kecl.ntt.co.jp/people/kaneko.takuhiro/projects/gac/index.html T. Kaneko et al., “Genera2ve A5ribute Controller with Condi2onal Filtered Genera2ve Adversarial Networks,” CVPR 2017. �

GAN �� ( �� ) �� StyleGAN2 [Karras+arXiv2019] �� CycleGAN [Zhu+ICCV2017] CFGAN [Kaneko+CVPR2017] ��

�� (1�� (��)��2 † �� † h$p://www.kecl.n$.co.jp/people/kaneko.takuhiro/#talks ��

��

�� 1�� ( Generator ) �� • �� 4 ��4 Pictures: A. Brock et al., “Large Scale GAN Training for High Fidelity Natural Image Synthesis,” ICLR 2019. ��

�� 1��5� • �� 1�� 1�� 5�� 1�� Pictures: A. Brock et al., “Large Scale GAN Training for High Fidelity Natural Image Synthesis,” ICLR 2019. ��

<latexit sha1_base64="Xnt9lX2cNEUMxaHM7bxRUcRtMvU=">AB9HicbVBNS8NAEJ3Ur1q/qh69LBahHixJFfQiFL14rGA/oI1ls920SzebuLsp1tDf4cWDIl79Md78N27bHLT1wcDjvRlm5nkRZ0rb9reVWVpeWV3Lruc2Nre2d/K7e3UVxpLQGgl5KJseVpQzQWuaU6bkaQ48DhteIPrid8YUqlYKO70KJugHuC+YxgbST3EV0i/z45cbFp+NOvmCX7CnQInFSUoAU1U7+q90NSRxQoQnHSrUcO9JugqVmhNxrh0rGmEywD3aMlTgCo3mR49RkdG6SI/lKaERlP190SCA6VGgWc6A6z7at6biP95rVj7F27CRBRrKshskR9zpEM0SQB1maRE85EhmEhmbkWkjyUm2uSUMyE48y8vknq5JyWyrdnhcpVGkcWDuAQiuDAOVTgBqpQAwIP8Ayv8GYNrRfr3fqYtWasdGYf/sD6/AHauJDa</latexit> <latexit sha1_base64="QLlkCjr28Ua3oZ3AMcbE39dZsPg=">AB73icbVBNSwMxEJ31s9avqkcvwSLUS9mtgl6EohePFewHtEvJptk2NJusSVasS/+EFw+KePXvePfmLZ70NYHA4/3ZpiZF8ScaeO6387S8srq2npuI7+5tb2zW9jb2iZKELrRHKpWgHWlDNB64YZTluxojgKOG0Gw+uJ3ygSjMp7swopn6E+4KFjGBjpdYTukRh6fGkWyi6ZXcKtEi8jBQhQ61b+Or0JEkiKgzhWOu258bGT7EyjHA6zncSTWNMhrhP25YKHFHtp9N7x+jYKj0USmVLGDRVf0+kONJ6FAW2M8JmoOe9ifif105MeOGnTMSJoYLMFoUJR0aiyfOoxQlho8swUQxeysiA6wMTaivA3Bm395kTQqZe+0XLk9K1avsjhycAhHUAIPzqEKN1CDOhDg8Ayv8ObcOy/Ou/Mxa1yspkD+APn8wd9o70</latexit> �� M��V�N�A�� AR , Flow , VAE , GAN • �M�G� ��M� ��M� �� 6��-��-��-��-1 �� 6��1��6�-��-� �-�-��6��-��-��1��-6�� (Chain Rule) �� F�� 1��-��-1 ��N x = f − 1 ( z ) z = f ( x ) Reference: I. Goodfellow, “NIPS 2016 Tutorial: GeneraBve Aversarial Networks,” NIPS 2016. �

�� c��M� AR , Flow , VAE , GAN • ��A��Nd ��F�� e�� i�� i�G �� 7��-��-��-��-1 �� 1��7��-��-� �-�-��-��-��1��-�� (Chain Rule) �N�i g��Va �� 1��-��-1 �b�� Reference: I. Goodfellow, “NIPS 2016 Tutorial: GeneraBve Aversarial Networks,” NIPS 2016. �

GAN 1/4 Generative Adversarial Networks [Goodfellow+2014] • – �� ( Generator ) �� ( Discriminator ) � �� D Random Real image G x Real/Fake Discriminator Latent code Fake image D ( x ) Generator z G ( z ) – Min-Max �� Generator vs. Discriminator � I. Goodfellow et al., “Generative Adversarial Nets,” NIPS 2014. ��

GAN 2/4 Genera&ve Adversarial Networks [Goodfellow+2014] • – �� ( Generator ) �� ( Discriminator ) � �� D Random Real image G x Real/Fake Discriminator Latent code Fake image D ( x ) Generator z G ( z ) �1��9�� – Min-Max �� Generator vs. Discriminator � Fake �� Discriminator � �� I. Goodfellow et al., “Generative Adversarial Nets,” NIPS 2014. ��

GAN 3/4 Genera&ve Adversarial Networks [Goodfellow+2014] • – �� ( Generator ) �� ( Discriminator ) � �� D Random Real image G x Real/Fake Discriminator Latent code Fake image D ( x ) Generator z G ( z ) Real/Fake �� – Min-Max �0�� Generator vs. Discriminator � Real/Fake �� Generator � �� 0� I. Goodfellow et al., “Generative Adversarial Nets,” NIPS 2014. ��

GAN 4/4 Genera&ve Adversarial Networks [Goodfellow+2014] • – �� ( Generator ) �� ( Discriminator ) � �� D Random Real image G x Real/Fake Discriminator Latent code Fake image D ( x ) Generator z G ( z ) �� 21 Real/Fake �� – Min-Max �� Generator vs. Discriminator � Generator � Discriminator � Min-Max �21 �� Generator �� I. Goodfellow et al., “Generative Adversarial Nets,” NIPS 2014. ��

��

Genera&ve Adversarial Networks NTT - PowerPoint PPT Presentation

Genera&ve Adversarial Networks NTT 2020 6 10 @JSAI2020 KS-02 AI MIRU2020

Adversarial Training Attacks on Deep Networks and Generative Adversarial Networks Erkut Erdem

Generative Adversarial Networks, Wasserstein Distance, and Adversarial Loss Zhiyu Min Alibaba

Automa'c Genera'on Control Using Ar'ficial Neural Networks By-

Neglected topics CS 446 Adversarial examples and deep networks 1 / 23 Adversarial

LAB MEETING: A Connection Between Generative Adversarial Networks, Inverse Reinforcement Learning

Introduction to Generative Adversarial Networks Ian Goodfellow, OpenAI Research Scientist NIPS

Robust Statistics and Generative Adversarial Networks Yuan YAO HKUST Chao Gao (Chicago) Jiyu

Genera&ve Stochas&c Networks Trainable by Backprop Yoshua

Generative Adversarial Networks presented by Ian Goodfellow presentation co-developed with Aaron

Generative Adversarial Networks Aaron Mishkin UBC MLRG 2018W2 1 Generative Adversial Networks

Adversarial Examples and Adversarial Training Ian Goodfellow, OpenAI Research Scientist Security

Adversarial Examples and Adversarial Training Ian Goodfellow, OpenAI Research Scientist Guest

Robust Statistics and Generative Adversarial Networks Yuan YAO HKUST 1 Chao GAO Jiyi LIU

CVPR 2020 Universal Adversarial Attacks Image agnostic and transferable across networks

Adversarial Examples and Adversarial Training Ian Goodfellow, OpenAI Research Scientist

ON THE ADVERSARIAL ROBUSTNESS OF UNCERTAINTY AWARE DEEP NEURAL NETWORKS APRIL 29 TH , 2019

Computing similarity between multiscale biological systems under uncertainty Kris Ghosh Miami

Variants and Combinations of Basic Models Stefano Ermon, Aditya Grover Stanford University

CS7015 (Deep Learning) : Lecture 23 Generative Adversarial Networks (GANs) Mitesh M. Khapra

The Wasserstein GAN Instructor: John Thickstun Discussion Board: Available on Ed Zoom Link:

The Role of Geographic Information in News Consumption Gebrekirstos G. Gebremeskel and Arjen P.

Learning-based Sampling over 3D Point Clouds presented by Dr. HOU Junhui, Assistant Professor

Program Analysis for Quantified Information Flow The 5th CREST Open Workshop Chunyan Mu joint

On the structure of polyharmonic Bergman spaces. Universidade de Lisboa, Instituto Superior T

Genera&ve Adversarial Networks NTT - PowerPoint PPT Presentation

Genera&ve Adversarial Networks NTT 2020 6 10 @JSAI2020 KS-02 AI MIRU2020

Adversarial Training Attacks on Deep Networks and Generative Adversarial Networks Erkut Erdem

Generative Adversarial Networks, Wasserstein Distance, and Adversarial Loss Zhiyu Min Alibaba

Automa'c Genera'on Control Using Ar'ficial Neural Networks By-

Neglected topics CS 446 Adversarial examples and deep networks 1 / 23 Adversarial

LAB MEETING: A Connection Between Generative Adversarial Networks, Inverse Reinforcement Learning

Introduction to Generative Adversarial Networks Ian Goodfellow, OpenAI Research Scientist NIPS

Robust Statistics and Generative Adversarial Networks Yuan YAO HKUST Chao Gao (Chicago) Jiyu

Genera&amp;ve Stochas&amp;c Networks Trainable by Backprop Yoshua

Generative Adversarial Networks presented by Ian Goodfellow presentation co-developed with Aaron

Generative Adversarial Networks Aaron Mishkin UBC MLRG 2018W2 1 Generative Adversial Networks

Adversarial Examples and Adversarial Training Ian Goodfellow, OpenAI Research Scientist Security

Adversarial Examples and Adversarial Training Ian Goodfellow, OpenAI Research Scientist Guest

Robust Statistics and Generative Adversarial Networks Yuan YAO HKUST 1 Chao GAO Jiyi LIU

CVPR 2020 Universal Adversarial Attacks Image agnostic and transferable across networks

Adversarial Examples and Adversarial Training Ian Goodfellow, OpenAI Research Scientist

ON THE ADVERSARIAL ROBUSTNESS OF UNCERTAINTY AWARE DEEP NEURAL NETWORKS APRIL 29 TH , 2019

Computing similarity between multiscale biological systems under uncertainty Kris Ghosh Miami

Variants and Combinations of Basic Models Stefano Ermon, Aditya Grover Stanford University

CS7015 (Deep Learning) : Lecture 23 Generative Adversarial Networks (GANs) Mitesh M. Khapra

The Wasserstein GAN Instructor: John Thickstun Discussion Board: Available on Ed Zoom Link:

The Role of Geographic Information in News Consumption Gebrekirstos G. Gebremeskel and Arjen P.

Learning-based Sampling over 3D Point Clouds presented by Dr. HOU Junhui, Assistant Professor

Program Analysis for Quantified Information Flow The 5th CREST Open Workshop Chunyan Mu joint

On the structure of polyharmonic Bergman spaces. Universidade de Lisboa, Instituto Superior T

Genera&ve Stochas&c Networks Trainable by Backprop Yoshua