A Distributed Reinforcement Learning Approach for Solving Optimization Problems

Combinatorial optimization is the seeking for one or more optimal solutions in a well defined discrete problem space. The optimization methods are of great importance in practice, particularly in the engineering design process, the scientific experiments and the business decision-making. We are investigating in this paper a distributed reinforcement learning based approach for solving combinatorial optimization problems. We are particularly focusing on the bidimensional protein folding problem, an NP-complete problem that refers to predicting the bidimensional structure of a protein from its amino acid sequence, an important optimization problem within many fields including bioinformatics, biochemistry, molecular biology and medicine. Our model is based on a distributed Q−learning approach. The experimental evaluation of the proposed system has provided encouraging results, indicating the potential of our proposal.