I am a first-year Ph.D. student in Computer Science at Cornell University, where I am mainly advised by Thorsten Joachims. I am also working with Yusuke Narita (Yale University) and some tech companies in Tokyo for large-scale empirical studies. I recently completed my bachelor’s degree in Industrial Engineering and Economics at the Tokyo Institute of Technology.
My research lies at the intersection of machine learning and causal inference called counterfactual learning. I am interested in the counterfactual nature of logged bandit feedback and human behavior data obtained from interactive systems, and ways of using biased real-world datasets to assist safe and better decision making in the wild.
- Counterfactual Learning and Evaluation
- Off-Policy Evaluation (Bandit / Reinforcement Learning)
- Heterogeneous Treatment Effect Prediction
- Offline Recommender Learning and Evaluation
- Learning from Human Behavior Data
- Statistical Machine Learning
- Fairness in Ranking / Fair Machine Learning
- Cornell University (2021-2026)
- Ph.D. Student in the Department of Computer Science
- Research Topics: Counterfactual Learning, Learning from Human Behavior Data
- Tokyo Institute of Technology (2016-2021)
- B.Eng. in Industrial Engineering and Economics
- Research Topics: Counterfactual Learning, Causal Inference, Recommender Systems
Yuta Saito, Shunsuke Aihara, Megumi Matsutani, and Yusuke Narita.
Open Bandit Dataset and Pipeline: Towards Realistic and Reproducible Off-Policy Evaluation
Neural Information Processing Systems (NeurIPS2021) Datasets and Benchmarks Track.
[paper] [software] [public dataset]
Haruka Kiyohara, Yuta Saito, Tatsuya Matsuhiro, Yusuke Narita, Nobuyuki Shimizu, and Yasuo Yamamoto.
Doubly Robust Off-Policy Evaluation for Ranking Policies under the Cascade Behavior Model
International Conference on Web Search and Data Mining (WSDM2022). (Acceptance rate=20.2%)
Nathan Kallus, Yuta Saito, and Masatoshi Uehara.
Optimal Off-Policy Evaluation from Multiple Logging Policies
International Conference on Machine Learning (ICML2021). (Acceptance rate=21.5%)
Yuta Saito and Shota Yasui.
Counterfactual Cross-Validation: Stable Model Selection Procedure for Causal Inference Models.
International Conference on Machine Learning (ICML2020). (Acceptance rate=21.8%)
[paper] [code] [slides]
Yuta Saito, Suguru Yaginuma, Yuta Nishino, Hayato Sakata, and Kazuhide Nakata.
Unbiased Recommender Learning from Missing-Not-At-Random Implicit Feedback.
International Conference on Web Search and Data Mining (WSDM2020). (Acceptance rate=14.8%)
[paper] [code] [slides]
- 2021.10: “Open Bandit Dataset and Pipeline: Towards Realistic and Reproducible Off-Policy Evaluation” has been accepted at NeurIPS2021 Datasets and Benchmarks Track!
- 2021.10: “Doubly Robust Off-Policy Evaluation for Ranking Policies under the Cascade Behavior Model” has been accepted at WSDM2022!
- 2021.09: I and Thorsten made a tutorial presentation at RecSys2021. The tutorial website is available here!
- 2021.05: “Optimal Off-Policy Evaluation from Multiple Logging Policies” has been accepted at ICML2021.
- 2021.03: I made an invited talk about the Open Bandit Project at the Tiger Talk (RMIT University)
- 2021.02: We released awesome-offline-rl, which is a collection of research and review papers for offline reinforcement learning (offline rl).
- 2020.07: My solo paper: “Doubly Robust Estimator for Ranking Metrics with Post-Click Conversions” has been accepted at RecSys2020.