NeurIPS 2019
Sun Dec 8th through Sat the 14th, 2019 at Vancouver Convention Center
Paper ID:2373
Title:Data Cleansing for Models Trained with SGD

The paper addresses the problem of estimating the influence of a data point, for models trained with SGD, for non-convex losses. It then describes a data cleaning method based on this estimate. Reviewers found the theory solid, and the proposed data-cleaning method promising.