Probabilistic models whose properties are defined via a global Free energy (that often is actually a Lyapunov function), that is minimized via the model's dynamics, which can be its intrinsic dynamics, or some training procedure imposed on the model.

Often, these models can be analyzed via Statistical physics, as these systems tend to minimize Free energyes.

They appear in several models in Machine learning, such as:

• Boltzmann machines
• Hopfield networks for symmetric connections
• Graphical models. Applications to discrete Optimization

https://github.com/davidBelanger/SPEN