Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Double Oracle Neural Architecture Search for Game Theoretic Deep Learning Models

Aye Phyu Phyu Aung, Xinrun Wang, Ruiyu Wang, Hau Chan, Bo An, Xiaoli Li, J. Senthilnath

TL;DR

A new approach to train deep learning models using game theory concepts including Generative Adversarial Networks (GANs) and Adversarial Training (AT) where the authors deploy a double-oracle framework using best response oracles to Adversarial Neural Architecture Search and Adversarial Training algorithms.

Abstract

In this paper, we propose a new approach to train deep learning models using game theory concepts including Generative Adversarial Networks (GANs) and Adversarial Training (AT) where we deploy a double-oracle framework using best response oracles. GAN is essentially a two-player zero-sum game between the generator and the discriminator. The same concept can be applied to AT with attacker and classifier as players. Training these models is challenging as a pure Nash equilibrium may not exist and even finding the mixed Nash equilibrium is difficult as training algorithms for both GAN and AT have a large-scale strategy space. Extending our preliminary model DO-GAN, we propose the methods to apply the double oracle framework concept to Adversarial Neural Architecture Search (NAS for GAN) and Adversarial Training (NAS for AT) algorithms. We first generalize the players' strategies as the trained models of generator and discriminator from the best response oracles. We then compute the meta-strategies using a linear program. For scalability of the framework where multiple network models of best responses are stored in the memory, we prune the weakly-dominated players' strategies to keep the oracles from becoming intractable. Finally, we conduct experiments on MNIST, CIFAR-10 and TinyImageNet for DONAS-GAN. We also evaluate the robustness under FGSM and PGD attacks on CIFAR-10, SVHN and TinyImageNet for DONAS-AT. We show that all our variants have significant improvements in both subjective qualitative evaluation and quantitative metrics, compared with their respective base architectures.

Double Oracle Neural Architecture Search for Game Theoretic Deep Learning Models

TL;DR

A new approach to train deep learning models using game theory concepts including Generative Adversarial Networks (GANs) and Adversarial Training (AT) where the authors deploy a double-oracle framework using best response oracles to Adversarial Neural Architecture Search and Adversarial Training algorithms.

Abstract

In this paper, we propose a new approach to train deep learning models using game theory concepts including Generative Adversarial Networks (GANs) and Adversarial Training (AT) where we deploy a double-oracle framework using best response oracles. GAN is essentially a two-player zero-sum game between the generator and the discriminator. The same concept can be applied to AT with attacker and classifier as players. Training these models is challenging as a pure Nash equilibrium may not exist and even finding the mixed Nash equilibrium is difficult as training algorithms for both GAN and AT have a large-scale strategy space. Extending our preliminary model DO-GAN, we propose the methods to apply the double oracle framework concept to Adversarial Neural Architecture Search (NAS for GAN) and Adversarial Training (NAS for AT) algorithms. We first generalize the players' strategies as the trained models of generator and discriminator from the best response oracles. We then compute the meta-strategies using a linear program. For scalability of the framework where multiple network models of best responses are stored in the memory, we prune the weakly-dominated players' strategies to keep the oracles from becoming intractable. Finally, we conduct experiments on MNIST, CIFAR-10 and TinyImageNet for DONAS-GAN. We also evaluate the robustness under FGSM and PGD attacks on CIFAR-10, SVHN and TinyImageNet for DONAS-AT. We show that all our variants have significant improvements in both subjective qualitative evaluation and quantitative metrics, compared with their respective base architectures.
Paper Structure (17 sections, 6 equations, 7 figures, 3 tables, 4 algorithms)

This paper contains 17 sections, 6 equations, 7 figures, 3 tables, 4 algorithms.

Table of Contents

  1. Introduction
  2. Related Works
  3. GAN, AT and NAS
  4. Double Oracle Algorithm.
  5. Preliminaries
  6. Adversarial Training
  7. Double Oracle Framework for GANs
  8. DONAS for Game Theoretic Deep Learning
  9. General Framework for DONAS
  10. DONAS for GAN
  11. DONAS for AT
  12. Experiments
  13. Experiments on DONAS for GAN
  14. Experiments on DONAS for AT
  15. Conclusion
  16. ...and 2 more sections

Figures (7)

  • Figure 1: An illustration of the general framework for DONAS. Figure adapted from lanctot2017unified.
  • Figure 1: The search process of DONAS-GAN obtaining diverse architectures.
  • Figure 2: Linear CKA between layers of the individual searched networks of DONAS-GAN trained on TinyImageNet dataset.
  • Figure 2: Linear CKA between layers of individual searched generator networks by DONAS-GAN for TinyImageNet dataset.
  • Figure 3: Linear CKA averaging between the same layers of DONAS-GAN searched networks trained on TinyImageNet. Generators 0, 1 and 3 have different architectures.
  • ...and 2 more figures