Subscribe to the gold package and get unlimited access to Shamra Academy
Register a new userApproximation and convergence of GANs training: an SDE approach
95
0
0.0
(
0
)
Added by
Haoyang Cao
Publication date
2020
fields
Informatics Engineering
and research's language is
English
Ask ChatGPT about the research
No Arabic abstract
Generative adversarial networks (GANs) have enjoyed tremendous empirical successes, and research interest in the theoretical understanding of GANs training process is rapidly growing, especially for its evolution and convergence analysis. This paper establishes approximations, with precise error bound analysis, for the training of GANs under stochastic gradient algorithms (SGAs). The approximations are in the form of coupled stochastic differential equations (SDEs). The analysis of the SDEs and the associated invariant measures yields conditions for the convergence of GANs training. Further analysis of the invariant measure for the coupled SDEs gives rise to a fluctuation-dissipation relations (FDRs) for GANs, revealing the trade-off of the loss landscape between the generator and the discriminator and providing guidance for learning rate scheduling.
rate research
Read More
Adversarial training has gained great popularity as one of the most effective defenses for deep neural networks against adversarial perturbations on data points. Consequently, research interests have grown in understanding the convergence and robustness of adversarial training. This paper considers the min-max game of adversarial training by alternating stochastic gradient descent. It approximates the training process with a continuous-time stochastic-differential-equation (SDE). In particular, the error bound and convergence analysis is established. This SDE framework allows direct comparison between adversarial training and stochastic gradient descent; and confirms analytically the robustness of adversarial training from a (new) gradient-flow viewpoint. This analysis is then corroborated via numerical studies. To demonstrate the versatility of this SDE framework for algorithm design and parameter tuning, a stochastic control problem is formulated for learning rate adjustment, where the advantage of adaptive learning rate over fixed learning rate in terms of training loss is demonstrated through numerical experiments.
It is hard to train Recurrent Neural Network (RNN) with stable convergence and avoid gradient vanishing and exploding, as the weights in the recurrent unit are repeated from iteration to iteration. Moreover, RNN is sensitive to the initialization of weights and bias, which brings difficulty in the training phase. With the gradient-free feature and immunity to poor conditions, the Alternating Direction Method of Multipliers (ADMM) has become a promising algorithm to train neural networks beyond traditional stochastic gradient algorithms. However, ADMM could not be applied to train RNN directly since the state in the recurrent unit is repetitively updated over timesteps. Therefore, this work builds a new framework named ADMMiRNN upon the unfolded form of RNN to address the above challenges simultaneously and provides novel update rules and theoretical convergence analysis. We explicitly specify key update rules in the iterations of ADMMiRNN with deliberately constructed approximation techniques and solutions to each subproblem instead of vanilla ADMM. Numerical experiments are conducted on MNIST and text classification tasks, where ADMMiRNN achieves convergent results and outperforms compared baselines. Furthermore, ADMMiRNN trains RNN in a more stable way without gradient vanishing or exploding compared to the stochastic gradient algorithms. Source code has been available at https://github.com/TonyTangYu/ADMMiRNN.
Generative Adversarial Networks (GANs) are powerful generative models, but suffer from training instability. The recently proposed Wasserstein GAN (WGAN) makes progress toward stable training of GANs, but sometimes can still generate only low-quality samples or fail to converge. We find that these problems are often due to the use of weight clipping in WGAN to enforce a Lipschitz constraint on the critic, which can lead to undesired behavior. We propose an alternative to clipping weights: penalize the norm of gradient of the critic with respect to its input. Our proposed method performs better than standard WGAN and enables stable training of a wide variety of GAN architectures with almost no hyperparameter tuning, including 101-layer ResNets and language models over discrete data. We also achieve high quality generations on CIFAR-10 and LSUN bedrooms.
Generative adversarial networks (GANs) have attracted intense interest in the field of generative models. However, few investigations focusing either on the theoretical analysis or on algorithm design for the approximation ability of the generator of GANs have been reported. This paper will first theoretically analyze GANs approximation property. Similar to the universal approximation property of the fully connected neural networks with one hidden layer, we prove that the generator with the input latent variable in GANs can universally approximate the potential data distribution given the increasing hidden neurons. Furthermore, we propose an approach named stochastic data generation (SDG) to enhance GANsapproximation ability. Our approach is based on the simple idea of imposing randomness through data generation in GANs by a prior distribution on the conditional probability between the layers. SDG approach can be easily implemented by using the reparameterization trick. The experimental results on synthetic dataset verify the improved approximation ability obtained by this SDG approach. In the practical dataset, four GANs using SDG can also outperform the corresponding traditional GANs when the model architectures are smaller.
We provide theoretical convergence guarantees on training Generative Adversarial Networks (GANs) via SGD. We consider learning a target distribution modeled by a 1-layer Generator network with a non-linear activation function $phi(cdot)$ parametrized by a $d times d$ weight matrix $mathbf W_*$, i.e., $f_*(mathbf x) = phi(mathbf W_* mathbf x)$. Our main result is that by training the Generator together with a Discriminator according to the Stochastic Gradient Descent-Ascent iteration proposed by Goodfellow et al. yields a Generator distribution that approaches the target distribution of $f_*$. Specifically, we can learn the target distribution within total-variation distance $epsilon$ using $tilde O(d^2/epsilon^2)$ samples which is (near-)information theoretically optimal. Our results apply to a broad class of non-linear activation functions $phi$, including ReLUs and is enabled by a connection with truncated statistics and an appropriate design of the Discriminator network. Our approach relies on a bilevel optimization framework to show that vanilla SGDA works.
suggested questions
Log in to be able to interact and post comments
comments
Fetching comments
Sign in to be able to follow your search criteria
