9:00 AM - 10:40 AM
[4Rin1-41] A game operation learning model using human play data
Keywords:Game AI
Recently, many artificial intelligence (AI) researches conducted in the video game industry employ neural networks.
By using the human play data as "teacher" data of the neural network, it is expected to obtain NN models that can output human game operation with high reproducibility.
In this research, we employed LSTM and CNN as our learning models and evaluated our results by comparing the winning percentage and the behavior of the game character.
Experimental results show that both models outperform the random-action
model and the LSTM model learns human-like behaviors, while the
CNN model presents patterned behaviors.
By using the human play data as "teacher" data of the neural network, it is expected to obtain NN models that can output human game operation with high reproducibility.
In this research, we employed LSTM and CNN as our learning models and evaluated our results by comparing the winning percentage and the behavior of the game character.
Experimental results show that both models outperform the random-action
model and the LSTM model learns human-like behaviors, while the
CNN model presents patterned behaviors.