A survey on generative adversarial networks: fundamentals and recent advances

Показать описание

The recent advances in generative modeling have gained a lot of attention both from ML researchers, practitioners, and businesses. Currently, the most attractive results are obtained using generative adversarial networks (GANs). Since their rise in 2014, GANs are being thoroughly studied from multiple viewpoints, e.g. probabilistic and game-theoretic ones.

This talk will survey several fundamental works related to the optimization objective of such models, and discuss the recently proposed strategies of their training.

00:00 Intro
03:44 SEMANTIC PROPERTIES
05:34 RANDOM NOISE
09:19 How TO COMPARE TWO DISTRIBUTIONS?
11:03 f-DIVERGENCES
13:07 CONVEX CONJUGATE
16:34 f-GAN
18:18 JS-GAN
21:01 NON-SATURATING GAN
23:49 INTEGRAL PROBABILITY METRICS
26:22 OPTIMAL TRANSPORT
30:02 WASSERSTEIN GAN
31:23 DISCRIMINATOR
34:17 GRADIENT PENALTIES
38:44 LIPSCHITZ NORM
41:52 SPECTRAL NORMALIZATION
44:57 RECEPTIVE FIELD
54:13 AUGMENTATIONS FOR GANS
55:56 ADAPTIVE DIFFERENTIABLE AUGMENTATIONS
56:38 GENERATOR
57:21 EMA OF WEIGHTS
01:02:17 PATH LENGTH REGULARIZATION
01:04:32 EVALUATION
01:09:46 FRECHET INCEPTION DISTANCE
01:11:55 KERNEL INCEPTION DISTANCE
01:18:37 PRECISION & RECALL
01:21:52 PERCEPTUAL PATH LENGTH
01:25:39 SPECTRAL ANALYSIS
01:33:40 OTHER METRICS
01:33:59 POSTPROCESSING
01:34:35 REJECTION SAMPLING
01:37:35 TRUNCATION TRICK
01:38:40 Discussion

[Chapters were auto-generated using our proprietary software - contact us if you are interested in access to the software]

The talk is based on the papers:
f-GAN: Training Generative Neural Samplers using Variational Divergence Minimization [NeurIPS 2016] Sebastian Nowozin, Botond Cseke, and Ryota Tomioka.

Which Training Methods for GANs do actually Converge? [ICML 2018] Lars Mescheder, Andreas Geiger, and Sebastian Nowozin.

Training generative adversarial networks with limited data [NeurIPS 2020] Tero Karras, Miika Aittala, Janne Hellsten, Samuli Laine, Jaakko Lehtinen, and Timo Aila.

Presenter BIO:

Denis Korzhenkov is a researcher at Samsung AI Center in Moscow. His work is mainly focused on the generative modeling of images. Denis serves as a reviewer at ICLR, CVPR, and ICCV. Got a MS in Mathematics at Lomonosov Moscow State University.

**This disclaimer informs readers that the views, thoughts, and opinions expressed in the talk belong solely to the author, and not necessarily to the author's employer, organization, committee, or other group or individual.**

-------------------------
Find us at:

Рекомендации по теме

Комментарии

Thank you! A great lecture if you want to dive into GANs.

guitaricet

Great video! Wanted to hear more about spectral analysis, which was really interesting. Too bad for the time sake

chaerinkong

Would love to hear any comment or input you have. Did you find the content interesting? Is there a place for improvement? Please leave any feedback you have here in a comment :)

ddai

A survey on generative adversarial networks: fundamentals and recent advances

A survey on generative adversarial networks: fundamentals and recent advances

A Comprehensive Survey of Generative Adversarial Networks GANs in Cybersecurity Intrusion Detection

A Comprehensive Survey of Generative Adversarial Networks GANs in Cybersecurity Intrusion Detection

Method for Exploring Generative Adversarial Networks (GANs) via Automatically Generated Image ...

12.11.2021 // AIJ // Master class on Generative Adversarial Networks. A.Maevskiy, HSE

L18.0: Introduction to Generative Adversarial Networks -- Lecture Overview

Generative Adversarial Networks - An introduction to GANs

DCGANs | Lecture 65 (Part 1) | Applied Deep Learning

Generative Adversarial Nets for Social Scientists

Generative Adversarial Networks (GANs)

Image Generation with Gans-based Techniques: A Survey

Generative Adversarial Networks

Generalization and Equilibrium in Generative Adversarial Nets (GANs)

StyleGAN3 [Explained]: Alias-Free Generative Adversarial Networks

[SAIF 2019] Day 1: New Perspectives on Generative Adversarial Networks - Simon Lacoste-Julien

12 Generative Adversarial Nets GANs - Machine Learning- Winter Term 20/21 - Freie Universität Berlin...

Rare Query Expansion Through Generative Adversarial Networks in Search Advertising

Yet Another Text Captcha Solver: A Generative Adversarial Network Based Approach

How to Modify Generative Adversarial Networks (GANs) Implementation for New Tasks and Losses

Generating Images with Generative Adversarial Networks

Generative Adversarial Networks - 11/14/2021

Generative Adversarial Imputation Network

Generative Adversarial Networks and TF-GAN (ML Tech Talks)

Part 1: Generative Adversarial Nets