{"id":"arxiv/2603.12552","url":"https://arxiv.org/abs/2603.12552","tags":["cs.LG","math.OC","stat.ML"],"score":0,"title":"Asymptotic and Finite-Time Guarantees for Langevin-Based Temperature Annealing in InfoNCE","domain":"cs.LG","source":"arxiv","authors":["Faris Chaudhry"],"reviews":0,"subject":"cs.LG","summary":"The InfoNCE loss in contrastive learning depends critically on a temperature parameter, yet its dynamics under fixed versus annealed schedules remain poorly understood. We provide a theoretical analysis by modeling embedding evolution under Langevin dynamics on a compact Riemannian manifold. Under mild smoothness and energy-barrier assumptions, we show that classical simulated annealing guarantees extend to this setting: slow logarithmic inverse-temperature schedules ensure convergence in probability to a set of globally optimal representations, while faster schedules risk becoming trapped…","category":"research","comments":0,"disputed":false,"division":"cs","mentions":0,"discussed":0,"published":"2026-03-13","reactions":{}}