Exhibition Program

Science of Machine Learning


Which is cause? Which is effect? Learn from data!

- Causal inference in time series via supervised learning -


Our goal is to automatically discover “causal relationships” from time series data, i.e., a sequence of data measured over time. Discovering causal relationships has key applications in various fields: e.g., finding that “R&D expenditure influences sales” is useful for decision making in companies; discovering gene regulatory relationships provides a key insight for drug discovery researches. To infer causal relationships, existing methods require us to select an appropriate mathematical expression (i.e., auto-regressive model) for each time series data, which is difficult without expertise in data analysis. For this problem, we build a novel approach that trains a machine learning model by using various data. Our method does not require a deep understanding of data analysis and therefore will help us to effectively make an important decision making in several situations.


  • [1] Y. Chikahara, A. Fujino, “Causal Inference in Time Series via Supervised Learning,” in Proc. 27th International Joint Conference on Artificial Intelligence (IJCAI), 2018.




Yoichi Chikahara, Learning and Intelligent Systems Research Group, Innovative Communication Laboratory