Head Pursuit: Probing Attention Specialization in Multimodal Transformers
Fuente:
arXiv
Salvato in:
| Autori principali: | , , , , |
|---|---|
| Natura: | Preprint |
| Pubblicazione: |
2025
|
| Soggetti: | |
| Accesso online: | |
| Tags: |
Aggiungi Tag
Nessun Tag, puoi essere il primo ad aggiungerne!!
|
| _version_ | 1866914255375171584 |
|---|---|
| author | Basile, Lorenzo Maiorca, Valentino Doimo, Diego Locatello, Francesco Cazzaniga, Alberto |
| author_facet | Basile, Lorenzo Maiorca, Valentino Doimo, Diego Locatello, Francesco Cazzaniga, Alberto |
| contents | Language and vision-language models have shown impressive performance across a wide range of tasks, but their internal mechanisms remain only partly understood. In this work, we study how individual attention heads in text-generative models specialize in specific semantic or visual attributes. Building on an established interpretability method, we reinterpret the practice of probing intermediate activations with the final decoding layer through the lens of signal processing. This lets us analyze multiple samples in a principled way and rank attention heads based on their relevance to target concepts. Our results show consistent patterns of specialization at the head level across both unimodal and multimodal transformers. Remarkably, we find that editing as few as 1% of the heads, selected using our method, can reliably suppress or enhance targeted concepts in the model output. We validate our approach on language tasks such as question answering and toxicity mitigation, as well as vision-language tasks including image classification and captioning. Our findings highlight an interpretable and controllable structure within attention layers, offering simple tools for understanding and editing large-scale generative models. |
| format | Preprint |
| id |
arxiv_https___arxiv_org_abs_2510_21518 |
| institution | arXiv |
| publishDate | 2025 |
| record_format | arxiv |
| spellingShingle | Head Pursuit: Probing Attention Specialization in Multimodal Transformers Basile, Lorenzo Maiorca, Valentino Doimo, Diego Locatello, Francesco Cazzaniga, Alberto Computer Vision and Pattern Recognition Computation and Language Machine Learning Language and vision-language models have shown impressive performance across a wide range of tasks, but their internal mechanisms remain only partly understood. In this work, we study how individual attention heads in text-generative models specialize in specific semantic or visual attributes. Building on an established interpretability method, we reinterpret the practice of probing intermediate activations with the final decoding layer through the lens of signal processing. This lets us analyze multiple samples in a principled way and rank attention heads based on their relevance to target concepts. Our results show consistent patterns of specialization at the head level across both unimodal and multimodal transformers. Remarkably, we find that editing as few as 1% of the heads, selected using our method, can reliably suppress or enhance targeted concepts in the model output. We validate our approach on language tasks such as question answering and toxicity mitigation, as well as vision-language tasks including image classification and captioning. Our findings highlight an interpretable and controllable structure within attention layers, offering simple tools for understanding and editing large-scale generative models. |
| title | Head Pursuit: Probing Attention Specialization in Multimodal Transformers |
| topic | Computer Vision and Pattern Recognition Computation and Language Machine Learning |
| url | https://arxiv.org/abs/2510.21518 |