Distributional Reinforcement Learning for Energy-Based Sequential Models

TL;DR

This paper introduces a distributional reinforcement learning approach for energy-based sequence models, enhancing their sampling and learning capabilities, demonstrated through experiments with global autoregressive models.

Contribution

It proposes a novel distributional RL method for sequential energy-based models, overcoming limitations of previous distillation techniques and broadening applicability.

Findings

Effective in improving sequence sampling

Applicable to various energy-based models

Demonstrated on GAM-based experiments

Abstract

Global Autoregressive Models (GAMs) are a recent proposal [Parshakova et al., CoNLL 2019] for exploiting global properties of sequences for data-efficient learning of seq2seq models. In the first phase of training, an Energy-Based model (EBM) over sequences is derived. This EBM has high representational power, but is unnormalized and cannot be directly exploited for sampling. To address this issue [Parshakova et al., CoNLL 2019] proposes a distillation technique, which can only be applied under limited conditions. By relating this problem to Policy Gradient techniques in RL, but in a \emph{distributional} rather than \emph{optimization} perspective, we propose a general approach applicable to any sequential EBM. Its effectiveness is illustrated on GAM-based experiments.