diffusion-based-neural-solvers

Introduces LoRe, a training-free wrapper that enforces per-step interaction budgets for iterative graph solvers, achieving substantial speedups and memory reductions on combinatorial optimization problems like MIS and TSP.