8 papers
The Geometry of Statistical Feature Learning in Mean-Field Langevin Dynamics
Zong Shang, Tomoya Wakayama, Guillaume Lecué +1
We introduce a geometric formulation of statistical feature learning for supervised regression. Feature learning is defined through a base--fiber decomposition: the base is the fea…
A Decision-Theoretic View of Test-Time Training: When, How Far, and Which Directions to Adapt
Tomoya Wakayama
Test-time training (TTT) adapts a pretrained model to each prompt via parameter updates, improving accuracy under pretraining-to-test distribution shifts. Yet, its performance ofte…
In-Context Learning Is Provably Bayesian Inference: A Generalization Theory for Meta-Learning
Tomoya Wakayama, Taiji Suzuki
This paper develops a finite-sample statistical theory for in-context learning (ICL), analyzed within a meta-learning framework that accommodates mixtures of diverse task types. We…
On Misspecified Error Distributions in Bayesian Functional Clustering: Consequences and Remedies
Fumiya Iwashige, Tomoya Wakayama, Shonosuke Sugasawa +1
Nonparametric Bayesian approaches provide a flexible framework for clustering without pre-specifying the number of groups, yet they are well known to overestimate the number of clu…
Ensemble Prediction via Covariate-dependent Stacking
Tomoya Wakayama, Shonosuke Sugasawa
This study proposes a novel approach to ensemble prediction, called "covariate-dependent stacking" (CDST). Unlike traditional stacking and model averaging methods, CDST allows mode…
Similarity-based Random Partition Distribution for Clustering Functional Data
Tomoya Wakayama, Shonosuke Sugasawa, Genya Kobayashi
Random partition distribution is a crucial tool for model-based clustering. This study advances the field of random partition in the context of functional spatial data, focusing on…