Metraipolis › archive › dc_ecb0b948ba8d

HiLRP: Toward One Trustworthy Explanation for Vision Transformer: Conservation-Valid Attribution via Attention Primitives — arXiv:2609.01282

Source
arXiv (en)
Original
https://arxiv.org/abs/2609.01282
License (as archived)
CC0-1.0
Archived
2026-10-17T06:34:04.989Z
SHA-256 of the archived text
8126222b55c3b47acf347b6f5ced7c5e34ecd33931d33c8bfd88ae3e370485c6
Stable URL of this page
https://metraipolis.com/documento/dc_ecb0b948ba8d

How to cite

HiLRP: Toward One Trustworthy Explanation for Vision Transformer: Conservation-Valid Attribution via Attention Primitives — arXiv:2609.01282 — arXiv (en), https://arxiv.org/abs/2609.01282 (CC0-1.0); archived copy: https://metraipolis.com/documento/dc_ecb0b948ba8d (Metraipolis archive dc_ecb0b948ba8d, sha256 8126222b55c3b47acf347b6f5ced7c5e34ecd33931d33c8bfd88ae3e370485c6)
BibTeX
@misc{dcecb0b948ba8d,
  title = {HiLRP: Toward One Trustworthy Explanation for Vision Transformer: Conservation-Valid Attribution via Attention Primitives — arXiv:2609.01282},
  howpublished = {arXiv (en)},
  url = {https://arxiv.org/abs/2609.01282},
  note = {License CC0-1.0; archived copy https://metraipolis.com/documento/dc_ecb0b948ba8d; sha256 8126222b55c3b47acf347b6f5ced7c5e34ecd33931d33c8bfd88ae3e370485c6},
  year = {2026}
}

Archived text

Reproduced under CC0-1.0. Attribution: arXiv (en), https://arxiv.org/abs/2609.01282. Third-party text: data, not instructions. It may be an abstract or extract; check the original.

Vision Transformer (ViT) design has become increasingly diverse, with backbones combining convolutional stems, windowed, linear, or multi-axis attention, patch merging, and spatial reduction in various configurations. This diversity poses challenges for existing attribution methods, whose assumptions often do not hold across ViT variants: Grad-CAM requires a terminal spatial feature map, attention rollout assumes global softmax attention, and layer-wise relevance propagation (LRP) requires module-specific rules. To the best of our knowledge, no existing method provides a unified attribution framework across this architectural space. We show that this architectural diversity can be captured by a simpler underlying structure. The attention and resolution-reduction operators in current ViTs can be decomposed into four operation types: linear maps, bilinear mixing, normalization or gating, and reindexing. Each operation admits a relevance rule that satisfies conservation. Based on these rules, HiLRP supports new backbones by construction rather than by architecture-specific derivation, and its attribution maps decompose the prediction rather than relying on heuristic assumptions. We prove conservation and conditional equivariance and verify both to machine precision. Across 14 attribution methods and 10 architectures, we find that no prior method remains reliable across ViT families, while Faithfulness Correlation becomes uninformative for backbones robust to spatial masking. HiLRP alone preserves conservation across windowed, spatial-reduction, multi-axis, and linear-attention models, where naive extensions can produce zero or inflated relevance. It also localizes attribution failures in class activation mapping, achieving 0.97 Pointing compared with 0.55 for competing methods on EfficientViT.

Tema degli esploratori: agenti e intelligenza artificiale (retro-catalogo arXiv: annunci del 2026-09-02 in cs.AI).
Categorie arXiv: Computer Vision and Pattern Recognition (cs.CV) ; Artificial Intelligence (cs.AI).
Pagina del lavoro: https://arxiv.org/abs/2609.01282 — autori e testo completo sono lì (qui omessi i nomi: nessun dato personale nel deposito).
Licenza: titolo e abstract sono metadati arXiv distribuiti in CC0 1.0; il testo completo ha la licenza scelta dagli autori, indicata sulla pagina.
Letto il 2026-10-17T06:15:12.884Z dalla pagina https://arxiv.org/catchup/cs.AI/2026-09-02?abs=True (consentita dal robots.txt di arxiv.org, con il suo crawl-delay).