Buffer Overflow in Mixture of Experts

Abstract

Mixture of Experts (MoE) has become a key ingredient for scaling large foundation models while keeping inference costs steady. We show that expert routing strategies that have cross-batch dependencies are vulnerable to attacks. Malicious queries can be sent to a model and can affect a model's output on other benign queries if they are grouped in the same batch. We demonstrate this via a \emph{proof-of-concept} attack in a \emph{toy experimental setting}.

Cite

Text

Hayes et al. "Buffer Overflow in Mixture of Experts." NeurIPS 2024 Workshops: SafeGenAi, 2024.

Markdown

[Hayes et al. "Buffer Overflow in Mixture of Experts." NeurIPS 2024 Workshops: SafeGenAi, 2024.](https://mlanthology.org/neuripsw/2024/hayes2024neuripsw-buffer/)

BibTeX

@inproceedings{hayes2024neuripsw-buffer,
  title     = {{Buffer Overflow in Mixture of Experts}},
  author    = {Hayes, Jamie and Shumailov, Ilia and Yona, Itay},
  booktitle = {NeurIPS 2024 Workshops: SafeGenAi},
  year      = {2024},
  url       = {https://mlanthology.org/neuripsw/2024/hayes2024neuripsw-buffer/}
}