Learning where to learn: Gradient sparsity in meta and continual learning

Abstract

Finding neural network weights that generalize well from small datasets is difficult. A promising approach is to learn a weight initialization such that a small number of weight changes results in low generalization error. We show that this form of meta-learning can be improved by letting the learning algorithm decide which weights to change, i.e., by learning where to learn. We find that patterned sparsity emerges from this process, with the pattern of sparsity varying on a problem-by-problem basis. This selective sparsity results in better generalization and less interference in a range of few-shot and continual learning problems. Moreover, we find that sparse learning also emerges in a more expressive model where learning rates are meta-learned. Our results shed light on an ongoing debate on whether meta-learning can discover adaptable features and suggest that learning by sparse gradient descent is a powerful inductive bias for meta-learning systems.

arXiv commentPublished at NeurIPS 2021

Cited by in corpus (5)

• Enhancing Few-Shot Image Classification through Learnable Multi-Scale Embedding and Attention Mechanisms
• When Meta-Learning Meets Online and Continual Learning: A Survey
• SIRe-Networks: Convolutional Neural Networks Architectural Extension for Information Preservation via Skip/Residual Connections and Interlaced Auto-Encoders
• Competitive plasticity to reduce the energetic costs of learning
• Group and Exclusive Sparse Regularization-based Continual Learning of CNNs