Download presentation
Presentation is loading. Please wait.
Published byBlaze Chandler Modified over 8 years ago
1
Generative Adversarial Nets ML Reading Group Xiao Lin Jul. 22 2015
2
I. Goodfellow, J. Pouget-Abadie, M. Mirza, B. Xu, D. Warde-Farley, S. Ozair, A. Courville and Y. Bengio. "Generative adversarial nets." NIPS, 2014.
3
Overview Problem: Generate adversarial examples Approach: Two player game Theory: Discriminative learning of distribution Potential application “Generative” Neural Nets Improving classification performance
4
Adversarial: a bit of background Visualizing HoG features “Why did my detector fail?” arXiv 2012 ICCV 2013 Visualizing CNNs arXiv 2013 ECCV 2014 CNNs can go wildly wrongarXiv 2013 Generative Adversarial NetsNIPS 2014 Google Deep Dream Facebook Eyescream CVPR 2015 ICLR 2015
5
A bit of background Visualizing HoG features “Why did my detector fail?” arXiv 2012 ICCV 2013 Visualizing CNNs arXiv 2013 ECCV 2014 CNNs can go wildly wrongarXiv 2013 Generative Adversarial NetsNIPS 2014 Google Deep Dream Facebook Eyescream CVPR 2015 ICLR 2015
6
A bit of background Visualizing HoG features “Why did my detector fail?” arXiv 2012 ICCV 2013 Visualizing CNNs arXiv 2013 ECCV 2014 CNNs can go wildly wrongarXiv 2013 Generative Adversarial NetsNIPS 2014 Google Deep Dream Facebook Eyescream CVPR 2015 ICLR 2015
7
A bit of background Visualizing HoG features “Why did my detector fail?” arXiv 2012 ICCV 2013 Visualizing CNNs arXiv 2013 ECCV 2014 CNNs can go wildly wrongarXiv 2013 Generative Adversarial NetsNIPS 2014 Google Deep Dream Facebook Eyescream CVPR 2015 ICLR 2015
8
A bit of background Visualizing HoG features “Why did my detector fail?” arXiv 2012 ICCV 2013 Visualizing CNNs arXiv 2013 ECCV 2014 CNNs can go wildly wrongarXiv 2013 Generative Adversarial NetsNIPS 2014 Google Deep Dream Facebook Eyescream CVPR 2015 ICLR 2015
9
A bit of background Visualizing HoG features “Why did my detector fail?” arXiv 2012 ICCV 2013 Visualizing CNNs arXiv 2013 ECCV 2014 CNNs can go wildly wrongarXiv 2013 Generative Adversarial NetsNIPS 2014 Google Deep Dream Facebook Eyescream CVPR 2015 ICLR 2015
10
A bit of background Visualizing HoG features “Why did my detector fail?” arXiv 2012 ICCV 2013 Visualizing CNNs arXiv 2013 ECCV 2014 CNNs can go wildly wrongarXiv 2013 Generative Adversarial NetsNIPS 2014 Google Deep Dream Facebook Eyescream CVPR 2015 ICLR 2015
11
Adversarial Framework Discriminative model “Police” Learns to determine whether a sample is from the model distribution of the generative model or the data distribution Generative model A team of “counterfeiters” trying to produce fake currency Try to fool the discriminative model with its model distribution Until the counterfeits are indistinguishable from the genuine articles Now the generative model generates a distribution indistinguishable from the data distribution
12
Related work (Table 2) Given examples. Learn model params Observe part of the example Infer the rest Generate examples according to model distribution Given example. Compute probability Design a model family with parameter θ
13
Flu Allergy Sinus Headache Nose=t Slide Credit: Dhruv Batra
14
Flu Allergy Sinus Headache Nose=t Slide Credit: Dhruv Batra
15
Generalized Denoising Autoencoder Y. Bengio, L. Yao, G. Alain and P. Vincent. "Generalized denoising auto-encoders as generative models." NIPS, 2013.
16
Approach: Objective
17
0/1=D(x)x=G(z) x 1: From data or 0: fake ones from G z, aka random noise Fake x
18
Approach: Optimization D G Data
19
Approach: Optimization D G Data Optimize D Improve G Eventually
20
Approach: Optimization
21
Approach: Convergence Best D given G: right in the middle of data and G
22
Approach: Convergence Best G: G=data and V=-log4
23
Approach: Convergence DG Data Optimize D Improve G Eventually
24
Results
37
Problems D must be in sync with G Train G to optimal => all output collapse to 1 point Tuning parameters “Enough capacity” Multi-mode distributions
38
Future work
Similar presentations
© 2025 SlidePlayer.com. Inc.
All rights reserved.