Mixing times and spectra of non-equilibrium symmetric exclusion processes on general graphs

What the math gives to ML

The paper identifies a non-reversible binary-particle dynamics whose spectrum is independent of the heat-bath occupation probabilities, even though those probabilities determine the non-equilibrium stationary distribution. This separates conditioning the stationary law from controlling relaxation, which could be useful in conditional stochastic neural modules where learned local probabilities should not unpredictably alter mixing. The most direct transfer is a graph-structured binary latent or refinement layer with learned heat-bath probabilities and fixed exchange/refresh rates. The extracted material does not include enough explicit constants from the mixing-time theorem to transfer its quantitative hitting-time bound reliably.