Deep Generative Models with Learnable Knowledge Constraints

TL;DR

This paper introduces a flexible, model-agnostic method to incorporate learnable, structured knowledge constraints into deep generative models by linking posterior regularization with reinforcement learning, improving generative quality.

Contribution

It establishes a mathematical link between posterior regularization and reinforcement learning, enabling the learning of constraints as rewards in DGMs.

Findings

Enhanced generative quality in human image synthesis

Improved templated sentence generation performance

Flexible application across various deep generative models

Abstract

The broad set of deep generative models (DGMs) has achieved remarkable advances. However, it is often difficult to incorporate rich structured domain knowledge with the end-to-end DGMs. Posterior regularization (PR) offers a principled framework to impose structured constraints on probabilistic models, but has limited applicability to the diverse DGMs that can lack a Bayesian formulation or even explicit density evaluation. PR also requires constraints to be fully specified a priori, which is impractical or suboptimal for complex knowledge with learnable uncertain parts. In this paper, we establish mathematical correspondence between PR and reinforcement learning (RL), and, based on the connection, expand PR to learn constraints as the extrinsic reward in RL. The resulting algorithm is model-agnostic to apply to any DGMs, and is flexible to adapt arbitrary constraints with the model jointly. Experiments on human image generation and templated sentence generation show models with learned knowledge constraints by our algorithm greatly improve over base generative models.