Population-based gradient descent weight learning for graph coloring problems

Graph coloring involves assigning colors to the vertices of a graph such that two linked by an edge receive different colors. problems are general models very useful formulate many relevant applications and, however, computationally difficult. In this work, population-based weight learning framework for solving is presented. Unlike existing methods specific considered problem, presented work targets generic objective introducing unified method can be applied problems. This distinguishes itself its approach formulates search solution as continuous tensor optimization problem and takes advantage gradient descent computed in parallel on graphics processing units. The proposed also characterized global loss function easily adapted usefulness demonstrated applying it solve typical performing extensive computational studies popular benchmarks. Improved best-known results (new upper bounds) equitable reported several large graphs.