Adaptive Divergence for Rapid Adversarial Optimization
arXiv:1912.00520 · doi:10.7717/peerj-cs.274
Abstract
Adversarial Optimization (AO) provides a reliable, practical way to match two implicitly defined distributions, one of which is usually represented by a sample of real data, and the other is defined by a generator. Typically, AO involves training of a high-capacity model on each step of the optimization. In this work, we consider computationally heavy generators, for which training of high-capacity models is associated with substantial computational costs. To address this problem, we introduce a novel family of divergences, which varies the capacity of the underlying model, and allows for a significant acceleration with respect to the number of samples drawn from the generator. We demonstrate the performance of the proposed divergences on several tasks, including tuning parameters of a physics simulator, namely, Pythia event generator.
References in corpus (22)
- Adam: A Method for Stochastic Optimization
- Very Deep Convolutional Networks for Large-Scale Image Recognition
- PYTHIA 6.4 Physics and Manual
- PYTHIA 6.2 Physics and Manual
- Unsupervised Representation Learning with Deep Convolutional Generative Adversarial Networks
- An Introduction to PYTHIA 8.2
- The ATLAS Simulation Infrastructure
- Progressive Growing of GANs for Improved Quality, Stability, and Variation
- DARTS: Differentiable Architecture Search
- Tuning PYTHIA 8.1: the Monash 2013 Tune
- Which Training Methods for GANs do actually Converge?
- Adversarially Learned Inference
- Wasserstein GAN
- Unrolled Generative Adversarial Networks
- The Cramer Distance as a Solution to Biased Wasserstein Gradients
- On Convergence and Stability of GANs
- Stabilizing Training of Generative Adversarial Networks through Regularization
- Amortised MAP Inference for Image Super-resolution
- On the Limitations of First-Order Approximation in GAN Dynamics
- Event generator tuning using Bayesian optimization
- Adversarial Variational Optimization of Non-Differentiable Simulators
- ChainGAN: A sequential approach to GANs