Motonobu Kanagawa
Brief Bio: I have been an Assistant Professor (Maitre de Conférences) in the Data Science Department at Eurecom since September 2019. I have also held a chair at 3IA Côte d’Azur since 2021. Previously, I was a research scientist with the Chair for the Methods of Machine Learning at the University of Tuebingen (Oct 2018 - Aug 2019) and at the Probabilistic Numerics Group in the Max Planck Institute for Intelligent Systems (Sep 2017 - Sep 2018), working with Prof. Philipp Hennig. Before this, I was a postdoc (Apr 2016 - Aug 2017) and a PhD. student (Apr 2013 - Mar 2016) at the Institute of Statistical Mathematics, working with Prof. Kenji Fukumizu.
Research interests: Statistics, Machine Learning and Simulation
Simulation is a fundamental methodology for understanding complex systems. Examples of such systems appear in a variety of fields, such as geophysical and environmental sciences (e.g., climate, weather, and natural disasters), social sciences (e.g., economics, finance, and insurance), and engineering (e.g., aviation engineering, traffic engineering, and architectural engineering), where simulation has been widely used. However, the reliability of a simulation depends on several factors, such as how accurately the underlying model (e.g., differential equations) can approximate the system of interest, and how accurately the execution of the simulation (e.g., a numerical solution to the differential equations) can approximate the model. For a simulation to be reliable, these factors must be systematically and objectively validated, but doing so manually is challenging.
Our objective is to develop statistical and machine learning methodologies for enhancing the reliability of simulation. As such methodologies themselves must also be reliable, we study mathematical theories to back the methodologies. Moreover, by cooperating with researchers and engineers from applied fields, we identify the needs in practice and develop tools that practitioners can easily use.
Contact Information
Eurecom, Data Science Department, Office 429
Campus SophiaTech, 450 Route des Chappes, 06410 Biot, France
Email: motonobu.kanagawa [CHAN] eurecom.fr (Please replace [CHAN] by @)
News
I have been awarded the Chris Daykin Prize from the International Actuarial Association for our paper "Intergenerational risk sharing in a defined contribution pension system: analysis with Bayesian optimization", together with An Chen (Ulm University) and Fangyuan Zhang (EDHEC Business School).
Our new paper "Variable Selection in Maximum Mean Discrepancy for Interpretable Distribution Comparison" is now available on arXiv (Nov 2023).
Our new paper "Comparing Scale Parameter Estimators for Gaussian Process Regression: Cross Validation and Maximum Likelihood'' is now available on arXiv (July 2023).
Our new paper ``When is Importance Weighting Correction Needed for Covariate Shift Adaptation?'' is now available on arXiv (Mar 2023).
Our paper ``Intergenerational Risk Sharing in a Defined Contribution Pension System: Analysis with Bayesian Optimization'' has been accepted by ASTIN Bulletin (March 2022)!
Our new paper ``Improved Random Features for Dot Product Kernels'' is now available on arXiv (Jan 2022).
We have written a new paper ``Intergenerational Risk Sharing in a Collective Defined-Contribution Pension System: A Simulation Study with Bayesian Optimization'' (June 2021).
Our paper "Counterfactual Mean Embeddings" has been accepted by JMLR! (June 2021)
A new preprint ``Connections and Equivalences between the Nyström Method and Sparse Variational Gaussian Processes'' is now available on arXiv (June 2021).
I have been awarded a new chair position at the 3IA Cote d'Azur (a French national research institute on AI) (June 2021).
We have significantly updated our preprint "Counterfactual Mean Embeddings", now containing several new results (Mar 2020).
We are organizing Workshop on Functional Inference and Machine Intelligence at EURECOM on 17 - 19 Feb 2020.
Our paper "Simulator Calibration under Covariate Shift with Kernels" has been accepted for publication at AISTATS 2020 (Jan 2020).
A new preprint "Simulator Calibration under Covariate Shift with Kernels" has now been available (Oct 2019).
Our paper ""Model-based Kernel Sum Rule: Kernel Bayesian Inference with Probabilistic Models" has been accepted for publication in Machine Learning (Oct 2019).
Our paper "Convergence Guarantees for Adaptive Bayesian Quadrature Methods" by Kanagawa and Hennig has been accepted for publication in NeurIPS 2019 (Sep 2019).
I have moved to Eurecom and started working as an Assistant Professor! (Sep 2019)
Our paper "On the positivity and magnitudes of Bayesian quadrature weights" by Karvonen, Kanagawa and Särkkä has been accepted for publication in Statistics and Computing (Aug 2018).
A new preprint "Convergence Guarantees for Adaptive Bayesian Quadrature Methods" by M. Kanagawa and P. Hennig (27 May 2019)
Our paper "Convergence Analysis of Deterministic Kernel-Based Quadrature Rules in Misspecified Settings" by Kanagawa, Sriperumbudur and Fukumizu has just been accepted for publication in Foundations of Computational Mathematics (30 Oct 2018)
Academic Services
Editorial Board Member of Journal of Machine Learning Research (JMLR)
Reviewing service:
TPAMI, Machine Learning, Statistics and Computing, JMLR, etc.
AISTATS(2016,2019, 2020), ICML (2015,2017,2018,2019,2020), NeuIPS (2015,2016,2017,2018,2019,2020)
Antibes, September 2019
Tuebingen, April 2019