Lifelong Learning with Sketched Structural Regularization

Haoran Li (Johns Hopkins University); Aditya R Krishnan (Johns Hopkins University); Jingfeng Wu (Johns Hopkins University); Soheil Kolouri (Vanderbilt University); Praveen Pilly (HRL Laboratories LLC); Vladimir Braverman (Johns Hopkins University)*


Preventing catastrophic forgetting while continually learning new tasks is an essential problem in lifelong learning. Structural regularization (SR) refers to a family of algorithms that mitigate catastrophic forgetting by penalizing the network for changing its "critical parameters" from previous tasks while learning a new one. The penalty is often induced via a quadratic regularizer defined by an importance matrix, e.g., the (empirical) Fisher information matrix in the Elastic Weight Consolidation framework. In practice and due to computational constraints, most SR methods crudely approximate the importance matrix by its diagonal. In this paper, we propose Sketched Structural Regularization (Sketched SR) as an alternative approach to compress the importance matrices used for regularizing in SR methods. Specifically, we apply linear sketching methods to better approximate the importance matrices in SR algorithms. We show that sketched SR: (i) is computationally efficient and straightforward to implement, (ii) provides an approximation error that is justified in theory, and (iii) is method oblivious by construction and can be adapted to any method that belongs to the SR class. We show that our proposed approach consistently improves various SR algorithms' performance on both synthetic experiments and benchmark continual learning tasks, including permuted-MNIST and CIFAR-100.