Donghwan Kim

Achieving the textbook $O(1/\sqrt{\epsilon})$ deterministic acceleration alongside optimal variance reduction is fundamentally impossible for stochastic monotone inclusions with fixed query budgets, overturning a standard assumption in non-convex min-max optimization.

Majority voting in LLM ensembles might be more about capability than diversity, with only 9.98% of subsets showing performance gains over the best model.

Judge signals from LLMs may not enhance optimization in table recognition, revealing a critical gap between evaluation ability and practical utility.