Wu and her colleagues hope to apply this strategy in the future to MILP problems that are even more complex, where collecting labeled data to train the model might be particularly challenging.
She suggests that they might be able to train the model on a smaller dataset and then modify it so that it can tackle a much larger optimization problem. The specialists are likewise keen on deciphering the learned model to more readily grasp the viability of various separator calculations.
Mathworks, the National Science Foundation (NSF), the MIT Amazon Science Hub, and the MIT Research Support Committee all provide partial funding for this study.