analiza aiqbench
This paper uses mechanistic interpretability to show that Vision Transformers encode spatial position in different ways depending on the embedding type: absolute embeddings store it as explicit features, while rotary embeddings rebuild it inside attention. The causal interventions, including a 24 to 31 point accuracy drop from permuting absolute positional features, make the findings more than descriptive. For readers building or debugging vision models, the practical takeaway is that the positional encoding choice changes where spatial information lives, which affects which interventions and probes will work.
How Vision Transformers Represent Position: A Mechanistic Comparison of Absolute and Rotary Positional Encodings
arXiv:2610.13992v1 Announce Type: cross Abstract: Positional embeddings encode the spatial position of each image patch in a Vision Transformer (ViT), allowing the model to distinguish different spatial arrangements of visual content. However, we know little about how the choice of embedding shapes a vision transformer's internal representation. In this work, we show that Absolute Positional Embeddings (APE) and Rotary Positional Embeddings (RoPE) encode spatial position through different internal mechanisms: APE stores position in explicit features, while RoPE builds it inside attention and rebuilds it when it is withheld. Specifically, in APE models, sparse autoencoders recover causal row- and column-specific features, whereas interventions on RoPE reveal distinct rotary components that separately encode row and column information. Additionally, causal interventions selectively disabling RoPE's row or column rotations show that decodable row and column information can be independently removed and, once rotations are restored, partly reconstructed, in part by axis-specialized attention heads. Both mechanisms are causally important for classification: permuting APE's positional features across patches reduces top-1 accuracy by 24--31 points, and ablating the rotations of RoPE's primary rebuilding head reduces it more than ablating those of any other head. Our findings show that the choice of positional encoding fundamentally changes how spatial information is organized and processed inside Vision Transformers. More broadly, our work demonstrates how mechanistic analysis can reveal how models construct, reconstruct, and maintain structured representations, providing a basis for targeted interventions in vision models.
Abstract:Positional embeddings encode the spatial position of each image patch in a Vision Transformer (ViT), allowing the model to distinguish different spatial arrangements of visual content. However, we know little about how the choice of embedding shapes a vision transformer's internal representation. In this work, we show that Absolute Positional Embeddings (APE) and Rotary Positional Embeddings (RoPE) encode spatial position through different internal mechanisms: APE stores position in explicit features, while RoPE builds it inside attention and rebuilds it when it is withheld. Specifically, in APE models, sparse autoencoders recover causal row- and column-specific features, whereas interventions on RoPE reveal distinct rotary components that separately encode row and column information. Additionally, causal interventions selectively disabling RoPE's row or column rotations show that decodable row and column information can be independently removed and, once rotations are restored, partly reconstructed, in part by axis-specialized attention heads. Both mechanisms are causally important for classification: permuting APE's positional features across patches reduces top-1 accuracy by 24--31 points, and ablating the rotations of RoPE's primary rebuilding head reduces it more than ablating those of any other head. Our findings show that the choice of positional encoding fundamentally changes how spatial information is organized and processed inside Vision Transformers. More broadly, our work demonstrates how mechanistic analysis can reveal how models construct, reconstruct, and maintain structured representations, providing a basis for targeted interventions in vision models.
Computer Vision and Pattern Recognition (cs.CV); Machine Learning (cs.LG)
arXiv:2610.13992 [cs.CV]
(or arXiv:2610.13992v1 [cs.CV] for this version)
https://doi.org/10.48550/arXiv.2610.13992 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.
Text integral sub licența deschisă a publicației · 330 cuvinte
