Uncertainty-based Continual Learning with Adaptive Regularization

TL;DR

This paper presents UCL, a novel uncertainty-based continual learning algorithm that reduces memory costs and improves task forgetting by leveraging a new interpretation of the KL divergence and node-wise uncertainty.

Contribution

UCL introduces a new interpretation of the KL divergence for Gaussian mean-field approximation, enabling efficient regularization with fewer parameters and better forgetting control.

Findings

UCL outperforms recent state-of-the-art methods on benchmarks.

It effectively balances stability and plasticity in continual learning.

UCL demonstrates strong performance in lifelong reinforcement learning tasks.

Abstract

We introduce a new neural network-based continual learning algorithm, dubbed as Uncertainty-regularized Continual Learning (UCL), which builds on traditional Bayesian online learning framework with variational inference. We focus on two significant drawbacks of the recently proposed regularization-based methods: a) considerable additional memory cost for determining the per-weight regularization strengths and b) the absence of gracefully forgetting scheme, which can prevent performance degradation in learning new tasks. In this paper, we show UCL can solve these two problems by introducing a fresh interpretation on the Kullback-Leibler (KL) divergence term of the variational lower bound for Gaussian mean-field approximation. Based on the interpretation, we propose the notion of node-wise uncertainty, which drastically reduces the number of additional parameters for implementing per-weight regularization. Moreover, we devise two additional regularization terms that enforce stability by freezing important parameters for past tasks and allow plasticity by controlling the actively learning parameters for a new task. Through extensive experiments, we show UCL convincingly outperforms most of recent state-of-the-art baselines not only on popular supervised learning benchmarks, but also on challenging lifelong reinforcement learning tasks. The source code of our algorithm is available at https://github.com/csm9493/UCL.