Fundamental thermodynamic limits on Bayesian inference

Understand and characterize the fundamental limits on energy and time required to perform Bayesian inference with physical systems governed by thermodynamic (Langevin) dynamics, establishing general lower bounds and optimality trade-offs.

Background

The paper frames thermodynamic Bayesian inference as an "entropy pump" that requires work to reduce uncertainty and dissipate heat. It analyzes dissipated work as a fundamental lower limit for Gaussian cases and discusses trade-offs between energy and time.

In the conclusion, the authors identify as an open and farther reaching goal the understanding of fundamental limits on Bayesian inference imposed by thermodynamics, particularly in terms of energy and time resources. Formalizing such limits would provide foundational guidance for hardware design and algorithm development.