Research Project
Our current research projects are as follows:
- Information Geometry in Machine Learning
- Theory and Application of Kernel Method
- Optimization-based Inference Project
- Statistical Analysis of Multimedia Data
- Machine Leaning for the Analysis of Neural Data
- Probabilistic Inference on Graphs
- Urban Intelligence Research
Project
1. Information Geometry in Machine Learning (Eguchi, Ikeda, Kobayashi, Fukumizu, Komori)
Information geometry is an approach to enhance geometrical methods toward a set of probabilistic models. In machine learning, the information geometry plays a fundamental role in elucidating probabilistic behaviors and statistical performance for proposed learning methods and algorithms to get better understanding for data. Thus such approaches to kernel methods, boosting and probabilistic coding have established by the use of conjugate linear connections and infinite dimensional exponential models. We challenge to deepen these results and to seek novel achievements in machine learning.
2. Theory and Application of Kernel Method (Fukumizu, Mochihashi, Kobayashi, Nishiyama)
This project aims to develop the new methodology for nonlinear data analysis using positive definite kernels or reproducing kernel Hilbert spaces. The project carries out researches on methods for analyzing dependence and causal relations among variables by extracting higher order moments of data in a computationally efficient manner, and a new Bayesian method with kernels. The research also includes necessary theoretical foundations and computational aspects.
3. Optimization-based Inference Project (Ito, Miyasato, Ikeda, Fushiki)
Computational inference is essential for understanding various phenomena from given data, and mathematical methodologies for computational inference are always demanded to be adaptable to changeable data both in quality and in quantity. The Optimization-based Inference Project focuses on optimization methodology as a fundamental tool for computational inference and aims to develop new inference techniques in statistical machine learning.
4. Statistical Analysis of Multimedia Data (Matsui, Mochihashi)
Achieving a safe, secure, and sustainable society
requires technology for utilizing massive amounts of
diverse data. We are developing statistical analysis
methods for discovering useful information from such data,
especially multimedia data of speech, music, images, and
text, in accordance with the objectives of classification
and prediction.
5. Machine Leaning for the Analysis of Neural Data (Koyama)
The recent progress in multiple-electrode recording makes it possible to study the simultaneous neural activity of many neurons. This allows us to understand how group of neurons act in synergy to define the function of a given brain region. In this project, we develop statistical methods for analyzing multiple neural data, in order to understand how the brain carries out information processing.
6. Probabilistic Inference on Graphs (Fukumizu, Ikeda)
This project carries out mathematical and theoretical
studies on the algorithms for probabilistic inference
based on graph structures, which are important in various
applications including Bayesian inference and decoding
algorithms. The main research topic is belief propagation,
which is an approximation algorithm propagating messages
on graphs. The project studies this algorithm from
viewpoints of information geometry, algebraic geometry,
algebraic topology, and graph theory, in addition to
analysis and further developments of such algorithms.
7. Urban Intelligence Research Project (Matsui)
The goal of this project is to promote urban resilience. We investigate theory and methodology based on statistics and machine learning for a wide range of subjects including situation analysis of energy/environment/ agriculture, risk management, security integration and design of urban resilience bond.