Same State, Different Task: Continual Reinforcement Learning without Interference

Abstract

Continual Learning (CL) considers the problem of training an agent sequentially on a set of tasks while seeking to retain performance on all previous tasks. A key challenge in CL is catastrophic forgetting, which arises when performance on a previously mastered task is reduced when learning a new task. While a variety of methods exist to combat forgetting, in some cases tasks are fundamentally incompatible with each other and thus cannot be learnt by a single policy. This can occur, in reinforcement learning (RL) when an agent may be rewarded for achieving different goals from the same observation. In this paper we formalize this "interference" as distinct from the problem of forgetting. We show that existing CL methods based on single neural network predictors with shared replay buffers fail in the presence of interference. Instead, we propose a simple method, OWL, to address this challenge. OWL learns a factorized policy, using shared feature extraction layers, but separate heads, each specializing on a new task. The separate heads in OWL are used to prevent interference. At test time, we formulate policy selection as a multi-armed bandit problem, and show it is possible to select the best policy for an unknown task using feedback from the environment. The use of bandit algorithms allows the OWL agent to constructively re-use different continually learnt policies at different times during an episode. We show in multiple RL environments that existing replay based CL methods fail, while OWL is able to achieve close to optimal performance when training sequentially.

Towards Continual Reinforcement Learning: A Review and Perspectives

“…Changepoint Detection: As discussed previously, one core issue in addressing settings with evolving task states is being able to detect change points or boundaries between significant switches without an oracle as in (Padakandla et al, 2019;Da Silva et al, 2006;Rosman and Ramamoorthy, 2012;Hadoux, Beynier, and Weng, 2014b;Li, Gu, Zhu, and Zhang, 2019;Kessler, Parker-Holder, Ball, Zohren, and Roberts, 2022;Luo, Jiang, Yu, Zhang, and Zhang, 2022). However, these approaches generally tend to be reactive to a changing distribution rather than proactive about anticipated changes in the future.…”

Transforming Pharma with Data Science, AI and Machine Learning

“…We followed Kessler et al (2022) in evaluating the efficacy of a continual learning method in forgetting, measuring stability and backward transfer, and forward transfer, also acting as a measure for plasticity. These measurements were recorded for each environment in the experiment suite T = (τ 1 , τ 2 , .…”

Minimum Description Length Control

“…In the sequential task setting (Moskovitz et al, 2022a;Pacchiano et al, 2022), tasks (MDPs) are sampled one at a time from P M , with the agent training on each until convergence. In contrast to continual learning (Kessler et al, 2021), the agent's goal is simply to learn a new policy for each task more quickly as more are sampled, rather than learning a single policy which maintains its performance across tasks. Another important setting is meta-RL, which we do not consider here.…”