Decomposing Query-Key Feature Interactions Using Contrastive Covariances

• URL: http://arxiv.org/abs/2602.04752v1
• Date: Wed, 04 Feb 2026 16:50:02 GMT
• Title: Decomposing Query-Key Feature Interactions Using Contrastive Covariances
• Authors: Andrew Lee, Yonatan Belinkov, Fernanda Viégas, Martin Wattenberg,
• Abstract summary: We study the query-key space -- the bilinear joint embedding space between queries and keys.<n>It is when features in keys and queries align in these low-rank subspaces that high attention scores are produced.
• Score: 75.38737409771085
• License: http://creativecommons.org/licenses/by-sa/4.0/
• Abstract: Despite the central role of attention heads in Transformers, we lack tools to understand why a model attends to a particular token. To address this, we study the query-key (QK) space -- the bilinear joint embedding space between queries and keys. We present a contrastive covariance method to decompose the QK space into low-rank, human-interpretable components. It is when features in keys and queries align in these low-rank subspaces that high attention scores are produced. We first study our method both analytically and empirically in a simplified setting. We then apply our method to large language models to identify human-interpretable QK subspaces for categorical semantic features and binding features. Finally, we demonstrate how attention scores can be attributed to our identified features.

Related papers

• Shortcut Invariance: Targeted Jacobian Regularization in Disentangled Latent Space [7.8904984750896885]
  Deep neural networks are prone to learning shortcuts, spurious and easily learned correlations.<n>We present a simple and effective training method that renders the classifier functionally invariant to shortcut signals.<n>We analyze this as targeted Jacobian regularization, which forces the classifier to ignore spurious features and rely on more complex, core semantic signals.
  arXiv  Detail & Related papers  (2025-11-24T07:09:08Z)
• Causal Attention with Lookahead Keys [52.63961482746826]
  In standard causal attention, each token's query, key, and value (QKV) are static and encode only preceding context.<n>We introduce CAuSal aTtention with Lookahead kEys (CASTLE), an attention mechanism that continually updates each token's keys as the context unfolds.
  arXiv  Detail & Related papers  (2025-09-09T00:15:23Z)
• Learning Structure-Supporting Dependencies via Keypoint Interactive Transformer for General Mammal Pose Estimation [24.010615027857007]
  We propose a Keypoint Interactive Transformer (KIT) to learn instance-level structure-supporting dependencies for general mammal pose estimation.<n>Our KITPose consists of two coupled components. The first component is to extract keypoint features and generate body part prompts.<n>Second, we propose a novel interactive transformer that takes feature slices as input tokens without performing spatial splitting.
  arXiv  Detail & Related papers  (2025-02-25T13:58:37Z)
• SCAPE: A Simple and Strong Category-Agnostic Pose Estimator [6.705257644513057]
  Category-Agnostic Pose Estimation (CAPE) aims to localize keypoints on an object of any category given few exemplars in an in-context manner. We introduce two key modules: a global keypoint feature perceptor to inject global semantic information into support keypoints, and a keypoint attention refiner to enhance inter-node correlation between keypoints. SCAPE outperforms prior arts by 2.2 and 1.3 PCK under 1-shot and 5-shot settings with faster inference speed and lighter model capacity.
  arXiv  Detail & Related papers  (2024-07-18T13:02:57Z)
• Open-Vocabulary Animal Keypoint Detection with Semantic-feature Matching [74.75284453828017]
  Open-Vocabulary Keypoint Detection (OVKD) task is innovatively designed to use text prompts for identifying arbitrary keypoints across any species. We have developed a novel framework named Open-Vocabulary Keypoint Detection with Semantic-feature Matching (KDSM) This framework combines vision and language models, creating an interplay between language features and local keypoint visual features.
  arXiv  Detail & Related papers  (2023-10-08T07:42:41Z)
• Learning Feature Matching via Matchable Keypoint-Assisted Graph Neural Network [52.29330138835208]
  Accurately matching local features between a pair of images is a challenging computer vision task. Previous studies typically use attention based graph neural networks (GNNs) with fully-connected graphs over keypoints within/across images. We propose MaKeGNN, a sparse attention-based GNN architecture which bypasses non-repeatable keypoints and leverages matchable ones to guide message passing.
  arXiv  Detail & Related papers  (2023-07-04T02:50:44Z)
• Attend to Who You Are: Supervising Self-Attention for Keypoint Detection and Instance-Aware Association [40.78849763751773]
  This paper presents a new method to solve keypoint detection and instance association by using Transformer. We propose a novel approach of supervising self-attention for multi-person keypoint detection and instance association.
  arXiv  Detail & Related papers  (2021-11-25T03:41:41Z)
• Compositional Attention: Disentangling Search and Retrieval [66.7108739597771]
  Multi-head, key-value attention is the backbone of the Transformer model and its variants. Standard attention heads learn a rigid mapping between search and retrieval. We propose a novel attention mechanism, called Compositional Attention, that replaces the standard head structure.
  arXiv  Detail & Related papers  (2021-10-18T15:47:38Z)
• The heads hypothesis: A unifying statistical approach towards understanding multi-headed attention in BERT [18.13834903235249]
  Multi-headed attention heads are a mainstay in transformer-based models. Different methods have been proposed to classify the role of each attention head based on the relations between tokens which have high pair-wise attention. We formalize a simple yet effective score that generalizes to all the roles of attention heads and employs hypothesis testing on this score for robust inference.
  arXiv  Detail & Related papers  (2021-01-22T14:10:59Z)
• Weakly-Supervised Aspect-Based Sentiment Analysis via Joint Aspect-Sentiment Topic Embedding [71.2260967797055]
  We propose a weakly-supervised approach for aspect-based sentiment analysis. We learn sentiment, aspect> joint topic embeddings in the word embedding space. We then use neural models to generalize the word-level discriminative information.
  arXiv  Detail & Related papers  (2020-10-13T21:33:24Z)

This list is automatically generated from the titles and abstracts of the papers in this site.

This site does not guarantee the quality of this site (including all information) and is not responsible for any consequences.