On January 30, the official account of the Journal of Computer Research and Development issued the article Recommend | Nanjing University Gao Yang Team Takes You to Play with the Scunty Egg, recommending the latest research achievement of Professor Gao Yang's team in the field of reinforcement learning, Solution of Scunty Egg Poker Game Based on Deep Reinforcement Learning.
Making decisions in complex environments with uncertain information is one of the difficulties that people often face in reality, so having the ability to make good decisions is considered one of the important abilities of artificial intelligence. Game type games, as a highly abstract representation of the real world, have the characteristics of well-defined and easy to test algorithm advantages and disadvantages, and have become the mainstream of research.
The article proposes a soft depth Monte Carlo SDMC method for the game of whipped egg poker. The SDMC method not only adopts a soft start method in the learning process, combined with existing strategies to accelerate the model training process, but also adopts soft action sampling. In actual combat, it ensures that the selected strategy samples actions without significant changes in the evaluation values of the current model, reducing the impact of variance during the training process and increasing the difficulty of being utilized by opponents.
The strategy model trained by the SDMC method won the championship in the 2nd China Artificial Intelligence Game Algorithm Competition. Through experimental comparison with the 1st champion strategy model and other strategy models in the 2nd edition, the effectiveness of this method in solving the game of whipped egg poker has been proven.
