Lifelong Learning with Sketched Structural Regularization
arXiv:2104.08604
Abstract
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 \emph{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 \emph{Sketched Structural Regularization} (Sketched SR) as an alternative approach to compress the importance matrices used for regularizing in SR methods. Specifically, we apply \emph{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 structural regularization 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.
References in corpus (9)
- Overcoming catastrophic forgetting in neural networks
- Improved Regularization of Convolutional Neural Networks with Cutout
- Wide Residual Networks
- Riemannian Walk for Incremental Learning: Understanding Forgetting and Intransigence
- Continual Learning Through Synaptic Intelligence
- Beyond Sharing Weights for Deep Domain Adaptation
- Continual Learning with Deep Generative Replay
- Less-forgetting Learning in Deep Neural Networks
- Empirical Analysis of the Hessian of Over-Parametrized Neural Networks