Constraints don't slow nonsmooth nonconvex optimization, Princeton-IAS paper shows
gautamcgoel · x · 2026-10-13
A new paper by Gon Buzaglo (Princeton) and Guy Kornowski (IAS) shows the oracle complexity of constrained nonsmooth nonconvex stochastic optimization is unchanged by constraints. Their projection-based and projection-free algorithms achieve optimal Õ(δ⁻¹ε⁻³) rates with stochastic gradients and Õ(dδ⁻¹ε⁻³) with zeroth-order queries, with a Frank-Wolfe-gap lower bound that holds even for smooth problems, via reduction to online linear optimization.
More from Research
- Softcatalà launches GarbellaVeus, a crowdsourced platform to validate Catalan speech datasets — jordiponsdotme · 2026-10-13
- NUS-Led Paper: Uncertainty-Guided Contraction Cuts Reasoning Post-Training Budget to 0.22% Without Losing Capability — NJU · 2026-10-13
- UCSD's SPLATIFY Turns 3DGS Papers into Trainable Code, Cutting Weeks of Reimplementation to Minutes — UCSanDiego · 2026-10-13
- TERRA turns spatial transcriptomics into virtual biology; cancer samples in, ~200M cells expected — mo_lotfollahi · 2026-10-13
- Embedding model turned into STT: M3 MacBook Air transcribes 1hr audio in 13s — TheMoonMidas · 2026-10-13
- Latent Personality Alignment: safer LLMs without any harmful training examples — burkov · 2026-10-13