[Re] Can gradient clipping mitigate label noise?

2021-01-31

ReScience C 2021

[Re] Can gradient clipping mitigate label noise?

David Mizrahi·Oğuz Kaan Yüksel·Aiday Marlen Kyzy

January 2021

Paper Code

Research context

Open in graph

Reproduction of gradient and loss clipping under label noise

A computational reproduction assessing the reported robustness behavior of gradient clipping and partially Huberized losses under symmetric label noise.

Research context

Uses

Studies

Produces

Reproduction of gradient and loss clipping under label noise

Related publications

Generalization Bounds for Autoregressive Processes and In-Context Learning

Semantic Perturbations with Normalizing Flows for Improved Generalization

Adversarial Training with Orthogonal Regularization