Kenji Fukumizu

About

Kenji Fukumizu is a leading figure in nonparametric Bayesian inference and kernel methods, whose work has fundamentally advanced how probabilistic models are learned from data without restrictive parametric assumptions. His research centers on kernel embedding of distributions—a powerful framework that represents probabilities in reproducing kernel Hilbert spaces, enabling elegant solutions to complex inference problems. Fukumizu’s major contributions include pioneering kernel Bayesian inference algorithms that combine the flexibility of nonparametric learning with the rigor of probabilistic graphical models. His highly cited work on Monte Carlo filtering using kernel embeddings (2014, 12 citations) and the kernel Monte Carlo filter (2015, 16 citations) revolutionized state-space model inference when observation models are unknown or only partially specified. More recently, his model-based kernel sum rule (2020) has extended these ideas to incorporate explicit probabilistic models, bridging the gap between data-driven and model-based approaches. Fukumizu’s research has had lasting impact on machine learning and signal processing, providing practitioners with principled tools for inference in high-dimensional, nonlinear systems where traditional methods fail.

Research Focus

Key Achievements

Top Papers

1. 1
   Filtering with State-Observation Examples via Kernel Monte Carlo Filter

   16 citations · 2015
2. 2
   Monte Carlo Filtering Using Kernel Embedding of Distributions

   12 citations · 2014
3. 3
   Model-based kernel sum rule: kernel Bayesian inference with probabilistic models

   2 citations · 2020

Key Collaborators

Contact & Links