Meta-analysis of Continual Learning
Published in Conference on Neural Information Processing Systems (NeurIPS) workshops, Vancouver, CA, 2019
Recommended citation: Cuong V. Nguyen, Alessandro Achille, Michael Lam, Tal Hassner, Vijay Mahadevan, Stefano Soatto. Meta-analysis of Continual Learning. Conference on Neural Information Processing Systems (NeurIPS) workshops, Vancouver, CA, 2019.
Abstract
We propose a novel meta-analysis to study the relationship between properties of task sequences and the performance of continual learning algorithms. Our analysis makes use of recent developments in task space modeling as well as correlation analysis to specify and analyze the properties we are interested in. As a case study, we apply our meta-analysis to study two properties of a task sequence: total complexity and sequential heterogeneity. The findings from our analysis suggest directions for improving continual learning benchmarks and methods.