Ching Lam CHOI

I'm an incoming MIT EECS PhD student at CSAIL, co-supervised by Phillip Isola, Antonio Torralba and Stefanie Jegelka.

During my CUHK bachelor's, I researched with wonderful people — Aaron Courville and Yann Dauphin at MILA; Wieland Brendel at Tübingen's Max Planck Institute; Serge Belongie at Copenhagen's Pioneer Centre; CUHK's Hongsheng Li (MMLab), Anthony Man-Cho So, Farzan Farnia and Qi Dou.

CV  /  Bio [WIP]  /  Coffee-hours  /  Blog [WIP]

Research

I'm interested in interpretability for diagnosing failures of robustness in ML. I research robust scaling — 1) better synthesis and curation of datasets; 2) designing sample-efficient, logic-symbolic models; 3) guaranteeing post-hoc fairness and privacy.

Gradient-based XAI fails to generalise from train to test data: regularisation crucially helps.

UAD is an adversarial examples game with a pure Nash equilibrium and better transferability.

BAKE distils self-knowledge of a network, by propagating samples similarities in each batch.

DivCo uses contrastive learning to reduce mode collapse in GANs.

News, Talks, Events

Website design credits to Jon Barron.