Ayaka Sakata
Ayaka Sakata applies statistical physics to learning and inference. She authored a 2018 method for estimating prediction error with approximate message passing and later studied Bayesian designs for group testing. Her work also includes generalized message-passing analysis of penalized regression and coauthored research on weight quantization. As of October 2026, she directs RIKEN’s Information Statistical Mechanics and Dynamics Team.
Timeline
-
2018
Estimating prediction error
Sakata proposed an estimator of prediction error using approximate message passing. The original paper describes its use with several sparse penalties and states conditions for asymptotic unbiasedness. It concerns how to evaluate the likely errors of a fitted statistical model.
-
2020–2021
Designing group tests
Her active-pooling paper uses Bayesian posterior prediction to design group tests. The preprint appeared in 2020 and the paper was published in 2021. This studies how information from previous tests can guide the selection of later pools.
-
2023–2024
Regression and quantized learning
She studied prediction errors for penalized regression using generalized approximate message passing in 2023. She also coauthored a 2024 study of weight quantization in learning models. These continue her analysis of inference and model behavior.
-
As of October 2026
Statistical physics for AI
Sakata directs RIKEN’s Information Statistical Mechanics and Dynamics Team. The team studies learning, inference and generation through statistical physics. The team aims for more efficient approximate inference and sampling, and stronger mathematical foundations for machine learning.
Sources and credits
Photo: www.riken.jp; creator not stated. License: Not stated by source. Source: riken.jp. Cropped to a square and resized. Ask us to remove or replace this photo.