Policy Iteration for Linear-Quadratic Stochastic Differential Games with State- and Control-Dependent Noise
arXiv:2608.17940
2026
Dynamics
2 ideas extracted · analyzed Sep 1, 2026
What the math gives to ML
The paper develops sequential policy iteration for stochastic differential games with state- and control-dependent noise. Its transferable mechanisms are preservation of mean-square stability during iteration, homotopy initialization from an easier low-diffusion problem, and a Frechet-derivative analysis showing that local convergence depends on noise sensitivity and update ordering. In neural networks, these mechanisms support a noise-continuation optimizer and an adaptive block-update schedule based on an estimated local spectral radius. Both proposals make quantitative predictions about the boundary between contracting and divergent training dynamics.
Ideas from this paper
△ Mechanism confirmed, baseline not beaten
2026
Train with a continuation parameter that gradually increases stochasticity, such as dropout, augmentation magnitude, gradient noise, or temperature, while monitoring the local mean-square stability of the parameter update. The network first solves a low-noise problem with a larger stability margin and is then continued toward the desired noisy objective instead of entering a high-noise regime abruptly.
Useful8/10
Difficulty5/10
Novelty6/10
△ Mechanism confirmed, baseline not beaten
2026
Replace simultaneous parameter updates with sequential block updates whose order is selected using estimated cross-block sensitivity. The paper shows that sequential policy updates can have a substantially smaller local contraction factor than decoupled or differently ordered updates; the neural analogue is to order attention, normalization, backbone, and head blocks according to the spectral radius of their composed update map.
Useful7/10
Difficulty6/10
Novelty7/10