This paper explores grokking in deep learning, linking delayed generalization to Softmax Collapse and proposing solutions to enable grokking without regularization through new activation functions and training algorithms.

https://arxiv.org/abs//2501.04697

YouTube: https://www.youtube.com/@ArxivPapers

TikTok: https://www.tiktok.com/@arxiv_papers

Apple Podcasts: https://podcasts.apple.com/us/podcast/arxiv-papers/id1692476016

Spotify: https://podcasters.spotify.com/pod/show/arxiv-papers