Inferring entropy production in many-body systems using nonequilibrium maximum entropy
A new AI-powered technique tackles the intractable problem of measuring disorder in complex systems like neural networks.
Researchers Miguel Aguilera, Sosuke Ito, and Artemy Kolchinsky developed a novel method using a nonequilibrium maximum entropy principle and convex duality. It infers entropy production (EP) in high-dimensional stochastic systems like many-body models and neural spike trains. Their approach uses only trajectory observables, avoiding complex probability reconstructions. They demonstrated it on a 1000-spin model and neural data, providing a hierarchical decomposition of EP with an intuitive 'thermodynamic uncertainty relation' interpretation.
Why It Matters
This provides a scalable tool to analyze irreversibility and energy dissipation in complex AI systems, biological networks, and materials science.