Fengxiang He

@vaishakbelle looks interesting!

Jessica Ciupa has a new paper accepted on "exploring verification frameworks for social choice alignment" at the Neurosymbolic Conference, 2025. This is one of the late-breaking papers. vaishakbelle.org/papers

🚨Just present our latest paper at #ICML2023 on the exciting topic of decentralized learning. Sad to miss Hawaii though!🏝️ 🎯TL; DR: Sounds like magic! We find that decentralized SGD implicitly performs sharpness-aware minimization! Paper: tinyurl.com/DSGD-SAM 🧵1/n

@dnnslmr Thanks Dennis :)

Very interesting also to connect this to another paper I recently read showing that modes in the loss surface correspond to different generalization strategies: x.com/dnnslmr/status… (3/3)

Want to learn how NLP models exploit spurious correlations instead of learning to solve the task? Read this fantastic survey by @DuMNCH, Fengxiang He, Na Zou, Dacheng Tao and Xia Hu! (1/3)

During graduation, Dean Goldbart usually gets to shake hands with each graduate as he wishes them well. Since our celebrations look a little different this year, he wanted to share this special sendoff for all our grads. #CNSGrad20 #UTGrad20 #FridayFeeling

In the past I’ve described this newspaper as a poor man’s Breitbart. It indulges racism, and uses it as part of its business model. This sort of commentary diminishes our society

The best criticisms of an argument suggest productive solutions to substantial flaws in the argument. Good criticisms attack the substance of the argument. Mediocre criticisms attack the style of the argument. Pathetic criticisms attack the identity of the arguer.