Fully Parameterized Quantile Function for Distributional Reinforcement Learning

TL;DR

This paper introduces a fully parameterized quantile function for distributional reinforcement learning, jointly training networks to better approximate true return distributions, leading to superior performance on Atari Games.

Contribution

It proposes a novel fully parameterized quantile function with a fraction proposal network and a quantile value network, improving distribution approximation in RL.

Findings

Outperforms existing distributional RL algorithms on Atari Games

Sets new record for non-distributed agents in Atari Learning Environment

Demonstrates significant performance improvements over prior methods

Abstract

Distributional Reinforcement Learning (RL) differs from traditional RL in that, rather than the expectation of total returns, it estimates distributions and has achieved state-of-the-art performance on Atari Games. The key challenge in practical distributional RL algorithms lies in how to parameterize estimated distributions so as to better approximate the true continuous distribution. Existing distributional RL algorithms parameterize either the probability side or the return value side of the distribution function, leaving the other side uniformly fixed as in C51, QR-DQN or randomly sampled as in IQN. In this paper, we propose fully parameterized quantile function that parameterizes both the quantile fraction axis (i.e., the x-axis) and the value axis (i.e., y-axis) for distributional RL. Our algorithm contains a fraction proposal network that generates a discrete set of quantile fractions and a quantile value network that gives corresponding quantile values. The two networks are jointly trained to find the best approximation of the true distribution. Experiments on 55 Atari Games show that our algorithm significantly outperforms existing distributional RL algorithms and creates a new record for the Atari Learning Environment for non-distributed agents.