Direct Optimization through $\arg \max$ for Discrete Variational Auto-Encoder

Abstract

Reparameterization of variational auto-encoders with continuous random variables is an effective method for reducing the variance of their gradient estimates. In the discrete case, one can perform reparametrization using the Gumbel-Max trick, but the resulting objective relies on an $\arg \max$ operation and is non-differentiable. In contrast to previous works which resort to softmax-based relaxations, we propose to optimize it directly by applying the direct loss minimization approach. Our proposal extends naturally to structured discrete latent variable models when evaluating the $\arg \max$ operation is tractable. We demonstrate empirically the effectiveness of the direct loss minimization technique in variational autoencoders with both unstructured and structured discrete latent variables.

Accepted by Neural Information Processing Systems (NeurIPS 2019)

References in corpus (20)

• Fashion-MNIST: a Novel Image Dataset for Benchmarking Machine Learning Algorithms
• Categorical Reparameterization with Gumbel-Softmax
• The Concrete Distribution: A Continuous Relaxation of Discrete Random Variables
• Junction Tree Variational Autoencoder for Molecular Graph Generation
• Attend, Infer, Repeat: Fast Scene Understanding with Generative Models
• Composing graphical models with neural networks for structured representations and fast inference
• REBAR: Low-variance, unbiased gradient estimates for discrete latent variable models
• Learning Latent Permutations with Gumbel-Sinkhorn Networks
• Variational inference for Monte Carlo objectives
• Grammar Variational Autoencoder
• Training Deep Neural Networks via Direct Loss Minimization
• Learning to Compose Words into Sentences with Reinforcement Learning
• Differentiable Perturb-and-Parse: Semi-Supervised Parsing with a Structured Variational Autoencoder
• Discrete Variational Autoencoders
• DVAE++: Discrete Variational Autoencoders with Overlapping Transformations
• ARM: Augment-REINFORCE-Merge Gradient for Stochastic Binary Networks
• MuProp: Unbiased Backpropagation for Stochastic Neural Networks
• DVAE#: Discrete Variational Autoencoders with Relaxed Boltzmann Priors
• On Sampling from the Gibbs Distribution with Random Maximum A-Posteriori Perturbations
• Local Expectation Gradients for Doubly Stochastic Variational Inference

Cited by in corpus (3)

• Improved Gradient-Based Optimization Over Discrete Distributions
• Joint Stochastic Approximation and Its Application to Learning Discrete Latent Variable Models
• Learning Randomly Perturbed Structured Predictors for Direct Loss Minimization