Neuron Constraints to Model Complex Real-World Problems

The benefits of combinatorial optimization techniques for the solution of real-world industrial problems are an acknowledged evidence; yet, the application of those approaches to many practical domains still encounters active resistance by practitioners, in large part due to the difficulty to come up with accurate declarative representations. We propose a simple and effective technique to bring hard-to-describe systems within the reach of Constraint Optimization methods; the goal is achieved by embedding into a combinatorial model a soft-computing paradigm, namely Neural Networks, properly trained before their insertion. The approach is flexible and easy to implement on top of available Constraint Solvers. To provide evidence for the viability of the proposed method, we tackle a thermal aware task allocation problem for a multi-core computing platform.