BODAME: Bilevel Optimization for Defense Against Model Extraction
We have published our paper in arXiv.
Yuto Mori, Atsushi Nitanda, and Akiko Takeda. BODAME: Bilevel Optimization for Defense Against Model Extraction. 2021. [arXiv]
As the title suggests, we formulate the problem of defending machine learning models from model extraction attacks as a bilevel optimization problem, and propose methods to solve.
In addition, I have published some of the survey materials that I had compiled when I was a master’s student.
It’s a survey that I’ve been compiling as a diary as we move into the coronavirus crisis. It mainly summarizes topics related to model extraction attacks, but also includes some abstracts of papers on the following topics.
- Active Learning
- Semi-supervised Learning
- Kernel Methods
- Machine Teaching
- Gaussian Process
- Poisoning
- Meta-Learning
