Hi all, I have a new result I'm excited to share. It turns out that *all existing implementations* of softmax and cross-entropy (that's JAX, PyTorch, textbooks, etc) have a subtle bug --- two bugs, actually. A simple rewrite fixes both, and eliminates "slingshots" when grokking:
井上裕太・河原林健一・宮下敦行・Bojan Mohar・Carsten Thomassen・Mikkel Thorupの論文“The Four Color Theorem with Linearly Many Reducible Configurations and Near-Linear Time Coloring”https://t.co/rjpYzEzumPが理論計算機科学のトップ会議FOCS2026のBest Paper Award受賞‼️おめでとう🎉🎉🎉
ICPC Announced that:
“Permits increase in gold-silver-bronze medals from 4-4-4 to 4-6-10 due to increased global growth.]”
Ref: https://t.co/HtJyFjBm9O