Motonobu Kanagawa

The Graduate University for Advanced Studies, SOKENDAI, University of Tübingen

Papers

Total Citations

H-Index

About

Motonobu Kanagawa is a researcher advancing the frontiers of nonparametric Bayesian inference and kernel methods, with a particular focus on state-space modeling and filtering. His work addresses a critical challenge: performing probabilistic inference when the observation model is unknown or cannot be specified parametrically. Kanagawa’s major contributions lie in developing novel Monte Carlo filtering algorithms that leverage kernel embeddings of distributions in reproducing kernel Hilbert spaces. His 2015 paper, “Filtering with State-Observation Examples via Kernel Monte Carlo Filter” (16 citations), and his foundational 2014 work, “Monte Carlo Filtering Using Kernel Embedding of Distributions” (12 citations), introduced principled nonparametric approaches to sequential state estimation, enabling inference without explicit probabilistic models. More recently, his 2020 paper on the “Model-based kernel sum rule” (2 citations) extends these ideas by combining kernel Bayesian inference with probabilistic models, creating a hybrid framework for nonparametric learning in graphical models. Though his citation counts are modest, Kanagawa’s work represents a technically rigorous and conceptually elegant bridge between classical Bayesian filtering and modern kernel methods, offering powerful tools for researchers working in robotics, signal processing, and machine learning where model uncertainty is a core challenge.