Learn-Prune-Share for Lifelong Learning
This work provides an incremental improvement to lifelong learning methods by simultaneously addressing multiple challenges for researchers and practitioners in sequential learning scenarios.
This paper proposes a learn-prune-share (LPS) algorithm for lifelong learning that addresses catastrophic forgetting, parsimony, and knowledge reuse. LPS partitions a neural network into task-specific components using an ADMM-based pruning strategy and integrates a selective knowledge-sharing scheme, consistently outperforming multiple state-of-the-art competitors on two benchmark datasets and a real-world radio frequency fingerprinting dataset.
In lifelong learning, we wish to maintain and update a model (e.g., a neural network classifier) in the presence of new classification tasks that arrive sequentially. In this paper, we propose a learn-prune-share (LPS) algorithm which addresses the challenges of catastrophic forgetting, parsimony, and knowledge reuse simultaneously. LPS splits the network into task-specific partitions via an ADMM-based pruning strategy. This leads to no forgetting, while maintaining parsimony. Moreover, LPS integrates a novel selective knowledge sharing scheme into this ADMM optimization framework. This enables adaptive knowledge sharing in an end-to-end fashion. Comprehensive experimental results on two lifelong learning benchmark datasets and a challenging real-world radio frequency fingerprinting dataset are provided to demonstrate the effectiveness of our approach. Our experiments show that LPS consistently outperforms multiple state-of-the-art competitors.