Parallel Counterfactual Regret Minimization in Crowdsourcing Imperfect-information Expanded Game
Jie Zhang, Kefan Li, Baoming Zhang, Ming Xu, Chongjun Wang · 2021
Counterfactual regret minimization (CFR) is one of the most widely used algorithms in iterative optimization algorithms. It is used to solve complex imperfect-information game problems. This paper introduced the Global Counterfactual Regret Minimization Local Update (GCFR+) to solve task planning problems in a crowdsourcing environment. We designed a parallel mechanism to alleviate possible parallel conflicts in actual crowdsourcing scenarios and increase personal rewards. First of all, we chose to test the performance of GCFR+ on data sets with different scales. Then we compared the result with the result of the decision model with a parallel mechanism. It can be seen that the parallel mechanism has significantly improved the efficiency of the decision model. Finally, unlike general CFR, we proved that GCFR+ is applicable to decision tree pruning of imperfect-information games.