NeurIPS 2019
Sun Dec 8th through Sat the 14th, 2019 at Vancouver Convention Center
Paper ID:7589
Title:Compacting, Picking and Growing for Unforgetting Continual Learning

The submission originally received mixed reviews, putting making it a borderline case. The reviewers appreciated the (relative) simplicity of the approach and the experimental evaluation of large-scale tasks. They also raised concerns about the originality of the approach, though, because all relevant components had already appeared in previous work, and they criticized a lack of mathematical derivations. After the author response, the reviewers discussed the work in detail. Ultimately, the decision was that the submission provides a valuable addition to the field by describing a well-engineered solution. Therefore, the work should be accepted. The authors are strongly encourage to make every possible effort to let other researchers benefit from their work, releasing their code and data, ideally pretrained models etc.