NeurIPS 2019
Sun Dec 8th through Sat the 14th, 2019 at Vancouver Convention Center
Paper ID:2956
Title:A Refined Margin Distribution Analysis for Forest Representation Learning


		
This paper formulates a forest representation learning approach (CASDF) as an additive model which boosts the augmented features and improves the upper bound on the generalization gap (by removing square root). The analysis results in a margin distribution reweighting scheme for deep forests. The paper is well-written, and is well-organized. Also the majority of reviewers find the contributions in the paper significant. However, R3 finds the experiments in the paper very limited. Nevertheless, given that the main emphasis and contribution of the paper is in theoretical analysis, I am fine with limited experimentation. Considering overall rating of this work, I recommend accept.