Publications
* denotes shared first authorship
2026
- ICLR
Scaling Attention via Feature SparsityInternational Conference on Learning Representations, 2026 - ICLR
CSRV2: Unlocking Ultra-Sparse EmbeddingsInternational Conference on Learning Representations, 2026
2025
- CVPR
Explaining Domain Shifts in Language: Concept erasing for Interpretable Image ClassificationProceedings of the Computer Vision and Pattern Recognition Conference, 2025 - Preprint
2024
- ACM MM
HICEScore: A Hierarchical Metric for Image Captioning EvaluationProceedings of the 32nd ACM International Conference on Multimedia, 2024 - Preprint
- Preprint