An Investigation Into Predict-and-Optimize Machine Learning

Predict-and-Optimize (PnO) is a relatively new machine learning paradigm that has attracted recent interest: it concerns the prediction of parameters that determine the value of solutions to an optimization problem, such that the optimizer ends up picking a good solution. Training estimators with standard loss functions like mean squared error...

Deepification: Learning Variable Ordering Heuristics in Constraint Optimization Problems

Constraint programming is a paradigm for solving combinatorial problems by checking whether constraints are satisfied in a constraint satisfaction problem or by optimizing an objective in a constraint optimization problem. To find solutions, the solver needs to find a variable and value ordering. Numerous heuristics designed by human experts...

Learning variable selection rules for the branch-and-bound algorithm using reinforcement learning

Mixed Integer Linear Programming (MILP) is a generalization of classical linear programming where we restrict some (or all) variables to take integer values. Numerous real-world problems can be modeled as MILPs, such as production planning, scheduling, network design optimization and many more. MILPs are, in fact, NP-hard. State-of-the-art...