Current browse context:
stat.ML
Change to browse by:
References & Citations
Statistics > Machine Learning
Title: Variational Continual Learning
(Submitted on 29 Oct 2017 (v1), revised 3 Nov 2017 (this version, v2), latest version 20 May 2018 (v3))
Abstract: This paper develops variational continual learning (VCL), a simple but general framework for continual learning that fuses online variational inference (VI) and recent advances in Monte Carlo VI for neural networks. The framework can successfully train both deep discriminative models and deep generative models in complex continual learning settings where existing tasks evolve over time and entirely new tasks emerge. Experimental results show that variational continual learning outperforms state-of-the-art continual learning methods on a variety of tasks, avoiding catastrophic forgetting in a fully automatic way.
Submission history
From: Cuong Nguyen [view email][v1] Sun, 29 Oct 2017 15:30:58 GMT (2102kb,D)
[v2] Fri, 3 Nov 2017 11:40:11 GMT (2103kb,D)
[v3] Sun, 20 May 2018 14:51:43 GMT (2258kb,D)
Link back to: arXiv, form interface, contact.