2 papers
cs.LG2026
Model Capacity Determines Grokking through Competing Memorisation and Generalisation Speeds
Yiding Song, Hanming Ye
Existing accounts of grokking explain the phenomena in terms of mechanistic frameworks such as circuit efficiency or lazy-to-rich transitions. However, despite a known dependence b…
cs.LG2025
Pruning Increases Orderedness in Recurrent Computation
Yiding Song
Inspired by the prevalence of recurrent circuits in biological brains, we investigate the degree to which directionality is a helpful inductive bias for artificial neural networks.…