mlsanityData to 14 Oct 2026

Debiasing CLIP: Interpreting and Correcting Bias in Attention Heads

arXiv 2505.17425 ↗First posted 23 May 2025cs.CV cs.CLcited by 4

Not on the map. Why?

Abstract (arXiv)

Multimodal models like CLIP have gained significant attention due to their remarkable zero-shot performance across various tasks. However, studies have revealed that CLIP can inadvertently learn spurious associations between target variables and confounding factors. To address this, we introduce textscLocate-Then-Correct (LTC), a contrastive framework that identifies spurious attention heads in Vision Transformers via mechanistic insights and mitigates them through targeted ablation. Furthermore, LTC identifies salient, task-relevant attention heads, enabling the integration of discriminative features through orthogonal projection to improve classification performance. We evaluate LTC on benchmarks with inherent background and gender biases, achieving over a >50% gain in worst-group accuracy compared to non-training post-hoc baselines. Additionally, we visualize the representation of selected heads and find that the presented interpretation corroborates our contrastive mechanism for identifying both spurious and salient attention heads. Code available at https://github.com/wj210/CLIPLTC.

Key facts. Cited by 4 papers in the map (data through 14 Oct 2026). Builds on 35 of its 42 references that are in the map. First posted on arXiv 23 May 2025.

Builds on

35 of 42 references.

References · 35

  1. 2017Attention Is All You Needcited by 60,361
  2. 2021Learning Transferable Visual Models From Natural Language Supervisioncited by 33,744
  3. 2020An Image is Worth 16x16 Words: Transformers for Image Recognition at Scalecited by 30,186
  4. 2017A Unified Approach to Interpreting Model Predictionscited by 6,555
  5. 2014Deep Learning Face Attributes in the Wildcited by 3,693
  6. 2019Distributionally Robust Neural Networks for Group Shifts: On the Importance of Regularization for Worst-Case Generalizationcited by 1,247
  7. 2022Toy Models of Superpositioncited by 971
  8. 2023Progress measures for grokking via mechanistic interpretabilitycited by 839
  9. 2021Robust fine-tuning of zero-shot modelscited by 797
  10. 2024MiniCPM: Unveiling the Potential of Small Language Models with Scalable Training Strategiescited by 723
  11. 2021Just Train Twice: Improving Group Robustness without Training Group Informationcited by 523
  12. 2022Visual Classification via Description from Large Language Modelscited by 336
23 more
  1. 2022Learning to Prompt for Open-Vocabulary Object Detection with Vision-Language Modelcited by 317
  2. 2022Language in a Bottle: Language Model Guided Concept Bottlenecks for Interpretable Image Classificationcited by 312
  3. 2024Long-CLIP: Unlocking the Long-Text Capability of CLIPcited by 300
  4. 2020Understanding the Role of Individual Units in a Deep Neural Networkcited by 255
  5. 2023Improving CLIP Training with Language Rewritescited by 217
  6. 2020No Subclass Left Behind: Fine-Grained Robustness in Coarse-Grained Classification Problemscited by 209
  7. 2023Interpreting CLIP's Image Representation via Text-Based Decompositioncited by 196
  8. 2020Causal Mediation Analysis for Interpreting Neural NLP: The Case of Gender Biascited by 191
  9. 2023Debiasing Vision-Language Models via Biased Promptscited by 149
  10. 2024Interpreting and Editing Vision-Language Representations to Mitigate Hallucinationscited by 104
  11. 2019FairFace: Face Attribute Dataset for Balanced Race, Gender, and Agecited by 87
  12. 2024A Multimodal Automated Interpretability Agentcited by 63
  13. 2022Contrastive Adapters for Foundation Model Group Robustnesscited by 62
  14. 2023Discovering and Mitigating Visual Biases through Keyword Explanationcited by 58
  15. 2023VisoGender: A dataset for benchmarking gender bias in image-text pronoun resolutioncited by 52
  16. 2023Mitigating Spurious Correlations in Multi-modal Models during Fine-tuningcited by 47
  17. 2024A Sober Look at the Robustness of CLIPs to Spurious Featurescited by 38
  18. 2024CLIP the Bias: How Useful is Balancing Data in Multimodal Learning?cited by 35
  19. 2024GenderBias-VL: Benchmarking Gender Bias in Vision Language Models via Counterfactual Probingcited by 31
  20. 2023Zero-Shot Robustification of Zero-Shot Modelscited by 24
  21. 2024Unknown Prompt, the only Lacuna: Unveiling CLIP's Potential for Open Domain Generalizationcited by 20
  22. 2024Towards Faithful Natural Language Explanations: A Study Using Activation Patching in Large Language Modelscited by 14
  23. 2024Gender Bias in Text-to-Video Generation Models: A case study of Soracited by 8

Built on by

Cited by year (2026 to 14 Oct)

Citing papers per year: 2025: 3; 2026: 1.

4 papers.

  1. 2025Enhancing Concept Localization in CLIP-based Concept Bottleneck Modelscited by 6
  2. 2025Steering Conceptual Bias via Transformer Latent-Subspace Activationcited by 1
  3. 2025A Multifaceted Analysis of Social Biases in Large Language Modelscited by 0
  4. 2026When Masking Helps or Hurts Robustness in Compressed CLIP: A Pre-Deployment Diagnosticcited by 0