Meta-learning with Auto-generated Tasks for Predicting Human Behaviour in Normal Form Games

Shaofei Chen, Jiaxing Chen, Peng Li, William Yuan, Zhen Zhen Hu

Abstract: In recent years, machine learning methods have been successfully applied to predict human behaviour in strategic settings. However, as available data of human behaviour is not always large enough and people's reasoning processes in different types of games are various, it is challenging to acquire a satisfied prediction model. In this paper, we employ a meta-learning method to improve learning performance in predicting human behaviour in normal form games. In particular, we first design a deep neural network that captures mixed human behaviour features to model and be learned to get a underlying behavioural predictor. Then, using a dataset of experimental human behaviour, we apply unsupervised learning to generate tasks and use meta-learning to improve the learning proficiency. Experimental results show that our proposed meta-learning method with the designed neural network and auto-generated tasks considerably increases the prediction accuracy and significantly exceeds the previous state-of-the-art.

Anonymous Url: I certify that there is no URL (e.g., github page) that could be used to find authors’ identity.

No Acknowledgement Section: I certify that there is no acknowledgement section in this submission for double blind review.

Code Of Ethics: I acknowledge that I and all co-authors of this work have read and commit to adhering to the ICLR Code of Ethics

Submission Guidelines: Yes

Please Choose The Closest Area That Your Submission Falls Into: Machine Learning for Sciences (eg biology, physics, health sciences, social sciences, climate/sustainability )