The applied mathematical aspects of statistical mechanics, a branch of physics, is pursued. The current main focus is on applications to the field of machine learning.
Overview of the research
Theoretical issues in information theory and machine learning are studied using techniques from statistical physics such as replica method, mean-field approximation, and Monte Carlo methods. Some notable research achievements so far include the clarification of reconstruction limits and the development of hyperparameter estimation methods in sparse estimation. Currently, I am interested in enhancing our understanding of modern machine learning models having complex structures, such as deep learning. To this end, theoretical analyses in some specific models and tasks are conducted. Some new learning frameworks are suggested based on those analyses.
Research Interests
数理生物
information theory
quantum dynamics
machine learning
learning theory
statistical physics
Research Areas
Informatics, Soft computing
Natural sciences, Mathematical physics and basic theory
Informatics, Statistical science
Papers
When resampling/reweighting improves feature learning in imbalanced classification?: A toy-model study
Tomoyuki Obuchi; Toshiyuki Tanaka
Transactions on Machine Learning Research, 22 Apr. 2025, Peer-reviewed
Analysis of high-dimensional Gaussian labeled–unlabeled mixture model via message-passing algorithm
Xiaosi Gu; Tomoyuki Obuchi
Journal of Statistical Mechanics: Theory and Experiment, 03 Mar. 2025, Peer-reviewed
Transfer Learning in $\ell_1$ Regularized Regression: Hyperparameter Selection Strategy based on Sharp Asymptotic Analysis
Koki Okajima; Tomoyuki Obuchi
Transactions on Machine Learning Research, 30 Jan. 2025, Peer-reviewed
Sparse Modeling for Spectrometer Based on Band Measurement
Kyoya Uemura; Tomoyuki Obuchi; Toshiyuki Tanaka
IEEE Transactions on Signal Processing, 25 Mar. 2024, Peer-reviewed
On Model Selection Consistency of Lasso for High-Dimensional Ising Models on Tree-like Graphs