NeurIPS 2019
Sun Dec 8th through Sat the 14th, 2019 at Vancouver Convention Center
Paper ID:6891
Title:Random Path Selection for Continual Learning

The reviewers in general find the proposed method interesting and the paper insightful. The rebuttal addressed several concerns raised in the original reviews, and the authors are requested to include the content of the rebuttal in the final version of the paper (e.g., the ablation study to assess the impact of progressively integrating individual components of RPS-Net, the effects of the network size and capacity, etc.). The authors should also address the remaining concerns of the reviewers, such as discussing the incremental nature of the work, removing unnecessarily exaggerated statements, presenting the learning algorithm more formally (e.g., what does it mean, "the skip connections and classification layer remains tunable in every case"?), etc.