Attention Sinks in VLMs Are Borrowed, Not Seen

Key points

• Visual attention sinks are image tokens that absorb a large share of attention over the image regardless of the query or task, and they mislead methods that treat attention maps as a record of where the model "looked.1
• Comparing six candidate detectors on ten VLMs from four families, we find that only the centered cosine similarity between an image token's key and the key of the language model's text-sink detects visual sinks.1
• Moving a sink's key component along this direction to the median of the image keys sends its attention back to the text-sink (Are Borrowed) while leaving model answers unchanged (Not Seen).1
• Building on this, we propose a training-free, entropy-triggered re-look that first cleans the attention map, then crops the most-attended region of the cleaned map and inserts it into the reasoning trace.1

Sources (1)

• [1]Attention Sinks in VLMs Are Borrowed, Not Seen
  arXiv (AI, ML, NLP, CV, robotics, multi-agent) · Oct 11, 07:20 PM

  We study how to detect and mitigate visual attention sinks in vision-language models (VLMs).

  Visual attention sinks are image tokens that absorb a large share of attention over the image regardless of the query or task, and they mislead methods that treat attention maps as a record of where the model "looked.

Extractive summary: sentences quoted from the sources.