Modeling a system'stemporal behaviour in reaction to external stimuli is a fundamentalprobleminmanyareas.

Thepresence of multiple structural states in the data represents a major bottleneck in existing processing pipelines, often requiring expert user supervision.

This makesthe performance ofNAS approaches inmore diverse areas poorly understood.

Wedemonstrate that this approach performs on par with auto-decoder basedapproaches whilebeinganorderofmagnitude fasterattest-time inference.

A common approach for solving BSPPs is by using first-order methods, where local gradient information is used to iteratively improve the solution in order to converge to an equilibrium asymptotically.

These have the opposite properties, with DRL having good sample efficiencyandpoor stability, while ESbeing vice versa. Recently,there havebeen attempts tocombine these algorithms, butthesemethods fullyrelyonsynchronous updatescheme, making it not ideal to maximize the benefits of the parallelism in ES.