graph-combinatorial-optimization

This paper introduces projection agents for graph combinatorial optimization using reinforcement learning and graph neural networks, operating in a continuous action embedding space to improve generalization and scalability, and releases the LaGCO-RL library.