Timezone: »

The Primacy Bias in Deep Reinforcement Learning
Evgenii Nikishin · Max Schwarzer · Pierluca D'Oro · Pierre-Luc Bacon · Aaron Courville

Tue Jul 19 03:30 PM -- 05:30 PM (PDT) @ Hall E #829

This work identifies a common flaw of deep reinforcement learning (RL) algorithms: a tendency to rely on early interactions and ignore useful evidence encountered later. Because of training on progressively growing datasets, deep RL agents incur a risk of overfitting to earlier experiences, negatively affecting the rest of the learning process. Inspired by cognitive science, we refer to this effect as the primacy bias. Through a series of experiments, we dissect the algorithmic aspects of deep RL that exacerbate this bias. We then propose a simple yet generally-applicable mechanism that tackles the primacy bias by periodically resetting a part of the agent. We apply this mechanism to algorithms in both discrete (Atari 100k) and continuous action (DeepMind Control Suite) domains, consistently improving their performance.

Author Information

Evgenii Nikishin (Mila, Université de Montréal)
Max Schwarzer (Mila, Google Brain)
Pierluca D'Oro (Mila, Université de Montréal)
Pierre-Luc Bacon (Mila)
Aaron Courville (Université de Montréal)

Related Events (a corresponding poster, oral, or spotlight)

More from the Same Authors