2 papers
cs.LG2026
Dimensional Criticality at Grokking Across MLPs and Transformers
Ping Wang
Abrupt transitions between distinct dynamical regimes are a hallmark of complex systems. Grokking in deep neural networks provides a striking example -- an abrupt transition from m…
cs.LG2026
Grokking as Dimensional Phase Transition in Neural Networks
Ping Wang
Neural network grokking -- the abrupt memorization-to-generalization transition -- challenges our understanding of learning dynamics. Through finite-size scaling of gradient avalan…