An agent-based approach to knapsack optimization problems

This paper proposes a new artificial intelligence framework that solves knapsack related problems (a class of complex combinatorial optimization problems). This framework, which is pseudo-parallel and stochastic, uses an agent based system to approximately solve the optimization problem. The system consists of active agents interacting in real time. They mimic the behavior of the parameters of the optimization problem while being individually driven by their own parameters, decision process, and fitness assessment. The application of the framework to the two-dimensional guillotine bin packing problem demonstrates its effectiveness both in terms of solution quality and run time.