Research
Interests
My research interests center around the statistical theory of machine learning algorithms, but I use the term algorithm broadly, to include the decisions made by the practioner interacting with data, receiving feedback, and subsequently tweaking algorithm code. I like to ask basic research questions that relate to solving learning problems in "hard" settings. In particular, I am interested in situations in which the learning algorithm (and its designer) have limited knowledge of the quality of data or feedback available, especially when our definition of "good performance" is something other than "a small average loss at test time".
Publications
Here is a list including most research-related publications and preprints written by me and my colleagues.
Making Robust Generalizers Less Rigid with Loss Concentration
Presented at ICONIP 2025, Onna, Japan.
Communications in Computer and Information Science 2754:391-405, 2026.
[proceedings, doi, arXiv, code]
Learning as choosing a loss distribution
VISxAI 2025, Vienna, Austria.
Soft ascent-descent as a stable and flexible alternative to flooding
Presented at NeurIPS 2024, Vancouver, Canada.
Advances in Neural Information Processing Systems 37:37955-37987, 2024.
[proceedings, doi, arXiv, code]
Criterion Collapse and Loss Distribution Control
Presented at ICML 2024, Vienna, Austria.
Proceedings of Machine Learning Research 235:18547-18567, 2024.
[proceedings, arXiv, code]
Robust variance-regularized risk minimization with concomitant scaling
Presented at AISTATS 2024, Valencia, Spain.
Proceedings of Machine Learning Research 238:1144-1152, 2024.
[proceedings, arXiv, code]
A Survey of Learning Criteria Going Beyond the Usual Risk
Journal: Journal of Artificial Intelligence Research, 78:781-821, 2023.
Oral: AAAI 2024 (Journal track), Vancouver, Canada.
Flexible risk design using bi-directional dispersion
Presented at AISTATS 2023, Valencia, Spain.
Proceedings of Machine Learning Research 206:1586-1623, 2023.
[proceedings, arXiv, code]
Learning with risks based on M-location
Journal: Machine Learning, 111:4679-4718, 2022.
Oral: ECML-PKDD 2022, Grenoble, France.
Spectral risk-based learning using unbounded losses
Presented at AISTATS 2022, online.
Proceedings of Machine Learning Research 151:1871-1886, 2022.
[proceedings, arXiv, code]
Anytime Guarantees under Heavy-Tailed Data
Presented at AAAI 2022, online.
Proceedings of the AAAI Conference on Artificial Intelligence, 36(6):6918-6925.
[proceedings, doi, arXiv, code]
Making learning more transparent using conformalized performance prediction
Presented at ICML 2021, Workshop on Distribution-Free Uncertainty Quantification.
[arXiv]
Learning with risk-averse feedback under potentially heavy tails
Presented at AISTATS 2021, online.
Proceedings of Machine Learning Research 130:892-900, 2021.
[proceedings, arXiv, code]
Robustness and scalability under heavy tails, without strong convexity
Presented at AISTATS 2021, online.
Proceedings of Machine Learning Research 130:865-873, 2021.
[proceedings, code]
Scaling-Up Robust Gradient Descent Techniques
Presented at AAAI 2021, online.
Proceedings of the AAAI Conference on Artificial Intelligence, 35(9):7694-7701.
[proceedings, doi, code]
Better scalability under potentially heavy-tailed feedback
Archival version.
PAC-Bayes under potentially heavy tails
Presented at NeurIPS 2019, Vancouver, Canada.
Advances in Neural Information Processing Systems 32, 2019.
[proceedings, arXiv]
Distribution-robust mean estimation via smoothed random perturbations
Preprint.
Better generalization with less data using robust gradient descent
Presented at ICML 2019, Long Beach, USA.
Proceedings of Machine Learning Research 97:2761-2770, 2019.
Robust gradient descent via back-propagation: A Chainer-based tutorial
Efficient learning with robust gradient descent
Journal: Machine Learning, 108(8):1523-1560, 2019.
Oral: ECML-PKDD 2019, Wurzburg, Germany.
Robust descent using smoothed multiplicative noise
Presented at AISTATS 2019, Naha, Japan.
Proceedings of Machine Learning Research 89:703-711, 2019.
Classification using margin pursuit
Presented at AISTATS 2019, Naha, Japan.
Proceedings of Machine Learning Research 89:712-720, 2019.
[proceedings, code]
Robust regression using biased objectives
Journal: Machine Learning, 106(9):1643-1679, 2017.
Oral: ECML-PKDD 2017, Skopje, North Macedonia.
Minimum proper loss estimators for parametric models
IEEE Transactions on Signal Processing, 64(3):704-713, 2016.
[doi]
Location robust estimation of predictive Weibull parameters in short-term wind speed forecasting
ICASSP 2015, Brisbane, Australia.
[doi]
Forecasting in wind energy applications with site-adaptive Weibull estimation
ICASSP 2014, Florence, Italy.
[doi]
Funding
I have been fortunate to receive many substantial grants as a principal investigator (PI) to support my work over the years, chiefly from JSPS (Japan Society for the Promotion of Science) and JST (Japan Science and Technology Agency). See their logos and links below.
For a complete list of the grants I've received as a PI, please see my researchmap page [en, ja]. This generous funding has been crucial to my progress, and has allowed me to efficiently run a diverse array of numerical experiments, travel to many international conferences to present my work, and ensure all my journal publications remain completely open-access.
