aiqbench analysis
This research introduces SAPER, a novel framework that uses spectral analysis of attention maps to identify and prune redundant heads in Vision Transformers without sacrificing accuracy. By leveraging interpretable insights into model architecture, the method achieves significant FLOPs reduction compared to existing baselines while maintaining strong performance on ImageNet-1K. This approach directly addresses the computational inefficiency and opacity inherent in large-scale foundation models like DINOv2.
Interpretability-Guided Soft Pruning of Attention Heads in Vision Transformers
arXiv:2608.00264v1 Announce Type: cross Abstract: Vision foundation models, such as DINOv2, learn highly expressive representations but rely on massive, opaque architectures that demand substantial computational power and memory. To provide an interpretable-guided and efficient solution to this issue, we first propose a spectral analysis and new visualization technique for individual attention heads based on the Laplacian eigenvectors of their attention maps. Building upon recent observations regarding the block structure of Vision Transformers, we perform semantic clustering of attention heads and identify functional redundancies. Leveraging these insights, we introduce SAPER (Soft Attention PrunER), an end-to-end differentiable pruning framework based on the LapSum Soft Top-K approach. Extensive experiments on ImageNet-1K demonstrate that SAPER achieves a highly favorable accuracy-efficiency trade-off, outperforming the competitive RAPTOR baseline in FLOPs reduction while preserving strong classification performance.
Abstract:Vision foundation models, such as DINOv2, learn highly expressive representations but rely on massive, opaque architectures that demand substantial computational power and memory. To provide an interpretable-guided and efficient solution to this issue, we first propose a spectral analysis and new visualization technique for individual attention heads based on the Laplacian eigenvectors of their attention maps. Building upon recent observations regarding the block structure of Vision Transformers, we perform semantic clustering of attention heads and identify functional redundancies. Leveraging these insights, we introduce SAPER (Soft Attention PrunER), an end-to-end differentiable pruning framework based on the LapSum Soft Top-K approach. Extensive experiments on ImageNet-1K demonstrate that SAPER achieves a highly favorable accuracy-efficiency trade-off, outperforming the competitive RAPTOR baseline in FLOPs reduction while preserving strong classification performance.
Computer Vision and Pattern Recognition (cs.CV); Artificial Intelligence (cs.AI)
arXiv:2608.00264 [cs.CV]
(or arXiv:2608.00264v1 [cs.CV] for this version)
https://doi.org/10.48550/arXiv.2608.00264 Focus to learn more arXiv-issued DOI via DataCite (pending registration)
arXivLabs is a framework that allows collaborators to develop and share new arXiv features directly on our website.
Both individuals and organizations that work with arXivLabs have embraced and accepted our values of openness, community, excellence, and user data privacy. arXiv is committed to these values and only works with partners that adhere to them.
Have an idea for a project that will add value for arXiv's community? Learn more about arXivLabs.
Full text under the publisher's open licence · 225 words
Related coverage
Humanity’s Last Invention — Richard Socher of Recursive
Latent Space · Sep 14, 2026
[AINews] High Taste Testers Ep. 1
Latent Space · Oct 17, 2026
[AINews] not much happened today
Latent Space · Sep 10, 2026
Observing Health Outcomes Using Remote Sensing Imagery and Geo-Context Guided Visual Transformer
arXiv · Sep 9, 2026