New AI Breakthrough Makes Machines Smarter, Faster, Cheaper

JANUS is a plug-and-play module that adds missing local geometric signals to population optimizers like CMA-ES, without replacing the host or requiring extra evaluations. It estimates a local Jacobian from recent evaluation traces, creating both a Gauss-Newton exploitation candidate and a trace-preserving exploration metric. Unlike MetaBBO methods, JANUS needs no offline training or task distribution—it learns geometry on the fly from the current run alone, while the host keeps full control. In same-protocol tests, JANUS improves CMA-ES on 11–15 of 16 BBOB functions across dimensions 30, 100, and 500, and achieves the best mean error on 13 of 16 functions at dimension 500 with no training cost. It also delivers a 936× geometric-mean improvement over the host on a 1000-dimensional BBOB subset. On structured tasks, JANUS achieves the best mean cost on 1135-dimensional UAV path planning, improving by 12.8% over the strongest baseline, and it improves multi-objective hosts on 12 of 38 tasks with zero significant regressions.