Solving constraint-satisfaction problems with distributed neocortical-like neuronal networks

TL;DR

This paper presents a neural network model inspired by neocortical motifs that can solve complex constraint satisfaction problems like Sudoku and graph coloring through unstable dynamics and inhibitory mechanisms.

Contribution

It introduces a biologically plausible network architecture capable of solving CSPs with mathematical guarantees of convergence.

Findings

Network can solve planar four-color graph coloring, Sudoku, and maximum independent set.

Unstable dynamics driven by recurrent excitation facilitate problem exploration.

Non-linear inhibitory constraints improve performance on hard problems.

Abstract

Finding actions that satisfy the constraints imposed by both external inputs and internal representations is central to decision making. We demonstrate that some important classes of constraint satisfaction problems (CSPs) can be solved by networks composed of homogeneous cooperative-competitive modules that have connectivity similar to motifs observed in the superficial layers of neocortex. The winner-take-all modules are sparsely coupled by programming neurons that embed the constraints onto the otherwise homogeneous modular computational substrate. We show rules that embed any instance of the CSPs planar four-color graph coloring, maximum independent set, and Sudoku on this substrate, and provide mathematical proofs that guarantee these graph coloring problems will convergence to a solution. The network is composed of non-saturating linear threshold neurons. Their lack of right saturation allows the overall network to explore the problem space driven through the unstable dynamics generated by recurrent excitation. The direction of exploration is steered by the constraint neurons. While many problems can be solved using only linear inhibitory constraints, network performance on hard problems benefits significantly when these negative constraints are implemented by non-linear multiplicative inhibition. Overall, our results demonstrate the importance of instability rather than stability in network computation, and also offer insight into the computational role of dual inhibitory mechanisms in neural circuits.