Entropy-regularised optimisation paper accepted in the Annals of Statistics
LesterMackey · x · 2026-11-05
Microsoft Research's Lester Mackey boosts a paper forthcoming in the Annals of Statistics. In optimisation, gradients guide convergence — but what's the analogue when the objective is over probability distributions? The paper introduces a way to evaluate sample approximations to optima for entropy-regularised objectives, with a core idea (local linearisation) simple enough to explain to a high schooler.
More from Research
- Paper on beliefs and coherence norms in LLMs published in Artificial Intelligence journal — mohitban47 · 2026-11-06
- New study traces late LM pretraining loss spikes to output-layer rounding errors — micahgoldblum · 2026-11-06
- Tomas Petricek traces how the very idea of a programming language was invented — round · 2026-11-06
- Treating Weights as Images: a Wild Idea to Generate Model Weights with Diffusion — zsakib_ · 2026-11-06
- Sanger Institute Director: AI-Accelerated Science Faces Institutional, Not Just Technical, Bottlenecks — ersatzben · 2026-11-06
- New Preprint Tests Whether RAG Helps LLMs Detect Implicit Misogyny — MaartenSap · 2026-11-06