
T-SKM-Net: Trainable Neural Network Framework for Linear Constraint Satisfaction via Sampling Kaczmarz-Motzkin Method
AAAI 2026Proceedings of the AAAI Conference on Artificial Intelligence, 40(17):14431–14439
A trainable Sampling Kaczmarz-Motzkin layer that enforces linear constraints on neural network outputs, with over 25× speedup over a standard solver on DC optimal power flow.