Jun Sakuma
Jun Sakuma works on the security and privacy of artificial intelligence. He directs RIKEN’s AI Security and Privacy Team. His team has worked on neural-architecture search, explanations of black-box models, adversarial examples and fairness constraints. He coauthored Privformer, a 2023 conference paper on privacy-preserving Transformers with multiparty computation. These methods address different risks in learning systems: data disclosure, manipulation of predictions and unequal decisions.
Timeline
-
2021
Neural-architecture search
Sakuma coauthored AdvantageNAS with Rei Sato and Youhei Akimoto, a method for efficient neural-architecture search using credit assignment.
-
2022
Explaining black-box models
Sakuma and collaborators studied discovering binary causal concepts with a variational autoencoder to explain black-box models.
-
2023
Privacy-preserving Transformers
Sakuma coauthored Privformer with Yoshimasa Akimoto, Kazuto Fukuchi and Yohei Akimoto. Privformer uses multiparty computation.
-
As of October 2026
AI security and privacy research
Sakuma directs RIKEN’s AI Security and Privacy Team. The team aims to protect private information and prevent malicious manipulation of AI. His work includes fairness-constrained regression and reversible adversarial examples.
Sources and credits
Photo: www.riken.jp; Jun Sakuma profile photograph; 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.