IBCL: Zero-shot Model Generation under Stability-Plasticity Trade-offs

Abstract: Algorithms that balance the stability-plasticity trade-off are well-studied in the continual learning literature. However, only a few of them focus on obtaining models for specified trade-off preferences. When solving the problem of continual learning under specific trade-offs (CLuST), state-of-the-art techniques leverage rehearsal-based learning, which requires retraining when a model corresponding to a new trade-off preference is requested. This is inefficient since there exist infinitely many different trade-offs, and a large number of models may be requested. As a response, we propose Imprecise Bayesian Continual Learning (IBCL), an algorithm that tackles CLuST efficiently. IBCL replaces retraining with constant-time convex combination. Given a new task, IBCL (1) updates the knowledge base in the form of a convex hull of model parameter distributions and (2) generates one Pareto-optimal model per given trade-off via convex combination without any additional training. That is, obtaining models corresponding to specified trade-offs via IBCL is zero-shot. Experiments whose baselines are current CLuST algorithms show that IBCL improves by at most 45% on average per task accuracy and by 43% on peak per task accuracy, while maintaining a near-zero to positive backward transfer. Moreover, its training overhead, measured by number of batch updates, remains constant at every task, regardless of the number of preferences requested.