Reproduce “Multi Attribute Bias Mitigation via Representation Learning” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Rajeev Ranjan Dwivedi, Ankur Kumar, Vinod K Kurmi
GMBM addresses bias in image classification by representing known cues during training and suppressing their influence on model updates. Its first stage combines image and bias features; its second penalizes gradient components along bias directions, with bias modules omitted at inference. On a COCO gender-classification split, its SBA is 0.10, versus 0.84 for ERM and 0.15 for BAdd; training requires known bias labels.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.