TY  - RPRT
TI  - FailSAE: Towards Interpretable Failure Prediction for Vision-Language Models via Sparse Autoencoders
AU  - Jie Ma
AU  - Zongxi Liu
AU  - Yi Zhu
PY  - 2026
UR  - https://arxiv.org/abs/2609.04276
ID  - 2609.04276
ER  -