Uncertainty-based Continual Learning with Adaptive Regularization
This work addresses memory efficiency and forgetting issues in continual learning, which is important for AI systems that learn sequentially, though it is incremental as it builds on existing Bayesian and regularization methods.
The paper tackles the problems of high memory cost and lack of graceful forgetting in regularization-based continual learning by introducing UCL, which uses node-wise uncertainty to reduce parameters and additional regularization terms, achieving state-of-the-art performance on supervised and reinforcement learning benchmarks.
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.