BBQ-Networks: Efficient Exploration in Deep Reinforcement Learning for Task-Oriented Dialogue Systems

TL;DR

This paper introduces BBQ-Networks, a novel exploration algorithm for deep reinforcement learning in dialogue systems that leverages Thompson sampling with Bayes-by-Backprop neural networks, achieving faster learning and improved efficiency.

Contribution

The paper proposes BBQ-Networks, an exploration method using Thompson sampling with Bayesian neural networks, enhancing learning speed in deep Q-learning for dialogue systems.

Findings

Faster learning compared to epsilon-greedy and other strategies.

Effective use of successful episodes to boost Q-learning.

Significant improvement in exploration efficiency.

Abstract

We present a new algorithm that significantly improves the efficiency of exploration for deep Q-learning agents in dialogue systems. Our agents explore via Thompson sampling, drawing Monte Carlo samples from a Bayes-by-Backprop neural network. Our algorithm learns much faster than common exploration strategies such as $\epsilon$-greedy, Boltzmann, bootstrapping, and intrinsic-reward-based ones. Additionally, we show that spiking the replay buffer with experiences from just a few successful episodes can make Q-learning feasible when it might otherwise fail.