Generative Adversarial Nets

Generative Adversarial Nets-ppt Download

  • Date:26 Aug 2020
  • Views:5
  • Downloads:0
  • Size:6.42 MB

Share Presentation : Generative Adversarial Nets

Download and Preview : Generative Adversarial Nets

Report CopyRight/DMCA Form For : Generative Adversarial Nets


Transcription:

Generative AdversarialML Reading GroupJul 22 2015 I Goodfellow J Pouget Abadie M Mirza B Xu D Warde Farley S Ozair A Courville and Y Bengio Generative adversarial nets NIPS .
Problem Generate adversarial examples Approach Two player game Theory Discriminative learning of distribution Potential application Generative Neural Nets.
Improving classification performance Adversarial a bit of backgroundVisualizing HoG features arXiv 2012 Why did my detector fail ICCV 2013arXiv 2013.
Visualizing CNNs ECCV 2014CNNs can go wildly wrong arXiv 2013Generative Adversarial Nets NIPS 2014Google Deep Dream CVPR 2015Facebook Eyescream ICLR 2015.
A bit of backgroundVisualizing HoG features arXiv 2012 Why did my detector fail ICCV 2013arXiv 2013Visualizing CNNs ECCV 2014.
CNNs can go wildly wrong arXiv 2013Generative Adversarial Nets NIPS 2014Google Deep Dream CVPR 2015Facebook Eyescream ICLR 2015 A bit of background.
Visualizing HoG features arXiv 2012 Why did my detector fail ICCV 2013arXiv 2013Visualizing CNNs ECCV 2014CNNs can go wildly wrong arXiv 2013.
Generative Adversarial Nets NIPS 2014Google Deep Dream CVPR 2015Facebook Eyescream ICLR 2015 A bit of backgroundVisualizing HoG features arXiv 2012.
Why did my detector fail ICCV 2013arXiv 2013Visualizing CNNs ECCV 2014CNNs can go wildly wrong arXiv 2013Generative Adversarial Nets NIPS 2014.
Google Deep Dream CVPR 2015Facebook Eyescream ICLR 2015 A bit of backgroundVisualizing HoG features arXiv 2012 Why did my detector fail ICCV 2013.
arXiv 2013Visualizing CNNs ECCV 2014CNNs can go wildly wrong arXiv 2013Generative Adversarial Nets NIPS 2014Google Deep Dream CVPR 2015.
Facebook Eyescream ICLR 2015 A bit of backgroundVisualizing HoG features arXiv 2012 Why did my detector fail ICCV 2013arXiv 2013.
Visualizing CNNs ECCV 2014CNNs can go wildly wrong arXiv 2013Generative Adversarial Nets NIPS 2014Google Deep Dream CVPR 2015Facebook Eyescream ICLR 2015.
A bit of backgroundVisualizing HoG features arXiv 2012 Why did my detector fail ICCV 2013arXiv 2013Visualizing CNNs ECCV 2014.
CNNs can go wildly wrong arXiv 2013Generative Adversarial Nets NIPS 2014Google Deep Dream CVPR 2015Facebook Eyescream ICLR 2015 Adversarial Framework.
Discriminative model Police Learns to determine whether a sample is from the model distribution of thegenerative 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 thedata distribution.
Related work Table 2 Given examples Learn model paramsObserve part of the exampleInfer the rest.
Generate examples accordingto model distributionGiven example Compute probabilityDesign a model family.
with parameter Flu AllergyHeadache Nose tSlide Credit Dhruv Batra Flu Allergy.
Headache Nose tSlide Credit Dhruv Batra GeneralizedAutoencoderY Bengio L Yao G Alain and P Vincent Generalized.
denoising auto encoders as generative models NIPS 2013 Approach Objective Approach Objective0 1 D x x G z 1 From data or.
0 fake ones from Gx z aka random noise Approach Optimization Approach OptimizationData Optimize D Improve G Eventually.
Approach Optimization Approach ConvergenceBest D given G right in the middle of data and G Approach ConvergenceBest G G data and V log4.
Approach ConvergenceData Optimize D Improve G Eventually 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 Future work.
This raised a lot of interest in vision community. Everyone hates CNNs beating our linear classifiers, and finally on these adversarial examples our linear classifiers outperform CNNs. On CVPR 14 there were a lot of discussions. The keynote speaker gave a pretty complicated mathematical theory to patch this.

Related Presentations