Every team makes mistakes:an initial report on predicting failure in teamwork

摘要: Voting among different agents is a powerful tool in problem solving, and it has been widely applied to improve the performance machine learning. However, potential of voting explored only improving ability finding correct answer complex problem. In this paper we present novel benefit voting, that not observed before: show can use patterns assess team predict their final outcome. This prediction be executed at any moment during problem-solving completely domain independent. We preliminary theoretical explanation why our method works, where accuracy better for diverse teams composed by than uniform made copies same agent. also perform experiments Computer Go domain, obtain high predicting outcome games. analyze 3 teams, works significantly team. Since approach independent, easily variety domains, such as video games Arcade Learning Environment.