New tuning-free optimizer speeds up large-scale AI training

Original: A simple adaptive accelerated optimistic dual averaging method for unconstrained convex optimization in $p$-norm geometry

Source: arxiv.org ↗