Hideaki Takahashi
Hello everyone,I recently graduated from the University of Tokyo with a Bachelor of Arts and Science in Informatics (class of 2024). My research interests revolve around AI, security, privacy, and systems.
During my undergraduate studies, I was fortunate to work under Prof. Alex Fukunaga, focusing on theoretical aspects of privacy-preserving path planning. I successfully published my senior thesis at AAMAS. Additionally, I spent a gap year as a research intern at Tsinghua University under the supervision of Prof. Yang Liu and Prof. Jingjing Liu, investigating privacy vulnerabilities in Federated Learning, resulting in a first-authored CVPR paper.
I am also passionate about open-source development. My project, AIJack, which aims to attack, defend, and debug machine learning models, has gained traction with over 300 stars on GitHub and has been widely used in research.
Education
- The University of Tokyo - Apr. 2019 ~ Mar. 2024
Bachelor of Arts and Science (Informatics). Major GPA 3.91/4.0 (Overall: 3.83/4.0)
Papers (peer-reviewed)
- [AAMAS'24] On the Transit Obfuscation Problem. Proceedings of the International Conference on Autonomous Agents and Multi-Agent Systems - 2024
Hideaki Takahashi*, Alex Fukunaga - [CVPR'23] Breaching FedMD, Image Recovery via Paired-Logits Inversion Attack. Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition - 2023
Hideaki Takahashi*, Jingjing Liu, and Yang Liu. - [ICLR'24] VFLAIR, A Research Library and Benchmark for Vertical Federated Learning. Proceedings of the International Conference on Learning Representations - 2024
Zou, Tianyuan, Zixuan Gu, Yuanqin He, Hideaki Takahashi, Yang Liu, Guangnan Ye and Ya-Qin Zhang
Preprints
- Eliminating Label Leakage in Tree‑based Vertical Federated Learning. arXiv:2307.10318 - 2023
Hideaki Takahashi*, Jingjing Liu, and Yang Liu - Difficulty of Detecting Overstated Dataset Size in Federated Learning. Technical Report of DPS, 10, Information Processing Society of Japan. - 2021
Hideaki Takahashi*, Kohei Ichikawa, and Keichi Takahashi
Softwares
- AIJack
AIJack is an easy-to-use open-source simulation tool for testing the security of your AI system against hijackers. It provides advanced security techniques like Differential Privacy, Homomorphic Encryption, K-anonymity, Debugging, and Federated Learning to guarantee protection for your AI. With AIJack, you can test and simulate defenses against attacks such as Evasion, Poisoning, Model Inversion, Backdoor, and Free-Rider. AIJack also provides a simple DBMS for SQL-based methods. - Gymbo
Gymbo is a Proof of Concept for a Gradient-based Symbolic Execution Engine implemented from scratch. Building on recent advancements that utilize gradient descent to solve SMT formulas, Gymbo leverages gradient descent to discover input values that fulfill each path constraint during symbolic execution. - My*
This series aims to implement various algorithms, compiliers and solvers from scratch.
- MyCompiler: Toy compiler from a simple language to LLVM-IR implemented from scratch in Haskell.
- MyPlanner: PDDL Solver implemented in C++ from scratch.
- MyOptimizer: Implementations of popular optimization and search algorithms.
Research Experience
- Institute for AI Industry Research, Tsinghua University - Federated Learning & Privacy Computing Intern
Jan. 2022 ‑ Feb. 2023
Conducted research on federated learning and privacy computing under the supervision of Prof. Yang Liu and Prof. Jingjing Liu. - Laboratory for Software Design and Analyis, Nara Institute of Science and Technology - Visiting Student
Aug. 2021 ‑ Sep. 2021
Conducted research on the free‑rider problem of federated learning under the supervision of Prof. Kohei Ichikawa and Prof. Keichi Takahashi.
Industry Experience
- Apple Inc. - Technical Internship: AIML/Software Engineer
Feb. 2024 - present
Working on AIML/software engineering. - UTokyo Economic Consulting Inc. - Research Assistant
Oct. 2020 - present
Working on research and social implementations of econometrics and machine learning. - RECRUIT - Data Science Intern
Aug. 2020 - Sep. 2020
Worked on a location‑based restaurant recommendation iOS app. - M3, Inc. - Data Analysis Intern
Feb. 2020 - Jun. 2020
Worked on a data analysis project in the field of medical surveys. - FRONTEO, Inc. - Research Intern
Sep. 2019 ‑ Mar. 2020
Worked on the detection of anomaly documents with NLP and network analysis.
Honors
- 45th/616 teams (silver medal), Kaggle, Google - Fast or Slow? Predict AI Model Runtime - 2023
Compiler Optimization, Automated Algorithmic Configuration - 67th/875 teams (bronze medal), Kaggle, Hungry Geese - 2021
AI Agent, Reinforcement Learning - 52nd/788 teams (bronze medal), Kaggle, Santa 2020 - 2021
AI Agent, Reinforcement Learning, Multi-Armed Bandit - 51st/1138 teams (silver medal), Kaggle, Google Research Football with Manchester City F.C. - 2020
AI Agent, Reinforcement Learning - 88th/1390 teams (bronze medal), Kaggle, Cornell Birdcall Identification - 2020
Audio Classification, Signal Processing