Join Our Paper Reading on Grokking Research from DeepMind!
Paper reading in 1-hour! 🚀 SallyAnn D. and Jason will discuss new research from DeepMind on grokking. The paper authors demonstrate two new and surprising behaviors: ungrokking, in which a network regresses from perfect to low test accuracy, and semi-grokking, in which a network shows delayed generalization to partial rather than perfect test accuracy. Here’s the paper: https://arxiv.org/abs/2309.02390 Join us! https://arize.com/resource/community-papers-reading/