A bi-stable neuronal model of Gibbs distribution

• A bi-stable neuron model is presented from which the Gibbs distribution can be derived.
• The model can be implemented in a stochastic (Boltzmann) machine composed of a system of bi-stable neurons.
• The model may have important implications in neural networks and Markov logic networks.

In this paper we present a bi-stable neuronal model consistent with the Gibbs distribution. Our approach utilizes a formalism used in stochastic (Boltzmann) machines with a bistable-neuron algorithm in which each neuron can exist in either an ON or an OFF state. The transition between the system’s states is composed of two random processes, the first one decides which state transition should be attempted and the second one decides if the transition is accepted or not. Our model can be easily extended to systems with asymmetrical weight matrices.