Search papers, labs, and topics across Lattice.
Affiliation:
4
0
5
Targeting evaluation compute at specific tail-sensitive transition kernels cuts policy CVaR estimation error by up to 76% compared to standard end-to-end rollouts.
Lean-based mathematical reasoning can be surprisingly unreliable, but this work shows how to selectively trust its judgments with provable risk bounds, paving the way for more robust automated math problem solving.
Forget heuristic knob-tuning: decoding algorithms are just regularized optimization problems hiding in plain sight.
