arXiv:2609.05808 · v2

Physical policy gradient theorem for in situ stochastic-adjoint training

首次提交 2026-09-05 09:53

版本更新 2026-11-07 07:24

雷达观测 2026-11-10 10:11

摘要

In situ adjoint training extracts parameter gradients directly from measurement, but has so far been limited to reciprocal or restricted systems. We introduce the physical counterpart of the policy gradient theorem: a stochastic-adjoint gradient estimator that lifts these constraints by trading reciprocity for nondegenerate diffusion. As validation, we train a nonlinear resonator network, whose own dynamics supply the policy, against antagonistic temporal modulations with gradients from measured stochastic trajectories alone, without finite differences or a separate adjoint experiment.

作者备注：8 pages, 2 figures, plus supplemental material (19 pages, 1 figure). Revised version