Date of this Version
We aim to solve the problem forming multiagent coalitions in uncertain environments where the coalition members’ capability of solving tasks change due to their learning. The MCFP-Mproblem for the agents refers to forming or joining coalitions on behalf of a set of human users so that those human users can solve tasks and improve their types (expertise) to improve their performances over time. MCFP-A problem for a set of agents refers to their forming or joining coalitions so that they are able to solve a set of assigned tasks while optimize their performance over time. We propose the Integrated Human Coalition Formation and Scaffolding (iHUCOFS) framework for solving MCFP-M. iHUCOFS agents balance the tradeoff between solving the current task well and improving the human users’ types to solve future tasks better by facilitating learning and teaching. We have veriﬁed iHUCOFS’ assumptions using simulation experiments and implemented the framework in ClassroomWiki–-a Wiki environment for collaborative learning. Our deployment results show that iHUCOFS’ agents can model the students accurately and form student groups to enhance collaboration and learning. We have proposed the Agents’ Dyadic Learning Inﬂuenced Tradeoff (ADLIT) framework that consists of a coalition formation protocol and approximation strategies to solve MCFP-A. ADLIT agents can form coalition to solve the current task well and improve their performance over time by improving their types with learning. Our empirical studies show that the ADLIT agents’ local learning interactions lead to a scalable and robust mechanism for improvement in the long term.