CoChat
Sign Up
Cited by 175Multimodal Machine Learning ApplicationsVisual Attention and Saliency DetectionAdvanced Neural Network Applications

Causal Attention for Vision-Language Tasks

Yang, Xu, Zhang, Hanwang, Qi, Guojun, Cai, Jianfei·2021·2021 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR)
DOI 10.1109/cvpr46437.2021.00972
Open this paper in CoChat — free account

Abstract

We present a novel attention mechanism: Causal Attention (CATT), to remove the ever-elusive confounding effect in existing attention-based vision-language models. This effect causes harmful bias that misleads the attention module to focus on the spurious correlations in training data, damaging the model generalization. As the confounder is unobserved in general, we use the front-door adjustment to realize the causal intervention, which does not require any knowledge on the confounder. Specifically, CATT is implemented as a combination of 1) In-Sample Attention (IS-ATT) and 2) Cross-Sample Attention (CS-ATT), where the latter forcibly brings other samples into every IS-ATT, mimicking the causal intervention. CATT abides by the QK-V convention and hence can replace any attention module such as top-down attention and self-attention in Transformers. CATT improves various popular attention-based vision-language models by considerable margins. In particular, we show that CATT has great potential in large-scale pre-training, e.g., it can promote the lighter LXMERT [57], which uses fewer data and less computational power, comparable to the heavier UNITER [14]. Code is published in https://github.com/yangxuntu/lxmertcatt.

Integrity signals

Registry

Crossref · 10.1109/cvpr46437.2021.00972

Cites

81 of 81 references verified0 citation errors · 0 not checkable

Statistics

Recomputing p-values needs the paper’s full text.

Claims the paper makes

This paper’s full text isn’t openly available, so its claims and self-check weren’t run. The registry and citation checks still apply.

Claims other papers make about this one

Each is checked against this paper’s own text: does it say what the citing paper says it does?

  • Overly compound pretraining objectives potentially upgrade the challenge of balancing among different loss terms, thus complicate the training optimization.

    “Meanwhile, overly compound pretraining objectives potentially upgrade the challenge of balancing among different loss terms, thus complicate the training optimization [199].”

    From Multimodal learning with transformers: A survey (2023) · cites this paper as [199]

  • Some recently successful practice used weakly aligned image-text pairs to perform pretraining and achieve both competitive performance and zero-shot learning capability for image classification, image-text retrieval, and open-ended visual question answering.

    “Some recently successful practice [9], [199], [205] used weakly aligned image-text pairs to perform pretraining, and achieve both competitive performance and zero-shot learning capability for image classification, image-text retrieval, and open-ended visual question answering, etc.”

    From Multimodal learning with transformers: A survey (2023) · cites this paper as [199]

  • Some recent attempts study the use of weakly-aligned cross-modal supervision to train Transformers to learn cross-modal interactions.

    “Some recent attempts [137], [199] study the use of weakly-aligned cross-modal supervision to train Transformers to learn the cross-modal interactions.”

    From Multimodal learning with transformers: A survey (2023) · cites this paper as [199]

  • Existing multimodal Transformer models are superior only for specific MML tasks, as they are designed specifically for only a subset of specific tasks.

    “existing multimodal Transformer models [9], [199], [263] are superior only for specific MML tasks, as they are designed specifically for only a subset of specific tasks [137], [142], [212], [249], [260], [261], [265], [266].”

    From Multimodal learning with transformers: A survey (2023) · cites this paper as [199]

  • As the learning scale expands exponentially, the training data become inevitably noisy and heterogeneous.

    “As the learning scale expands exponentially, the training data become inevitably noisy and heterogeneous [9], [199], [263].”

    From Multimodal learning with transformers: A survey (2023) · cites this paper as [199]

Cites

Flagged first, then by citations

2 of the 83 references couldn’t be matched to a known paper.

Cited by

Sorted by citations

Related

Authors
Yang, X., Zhang, H., Qi, G., & Cai, J.
Affiliations
  • Nanyang Technological University (SG)
  • Monash University (AU)
ORCID
Venue
2021 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR)
Publisher
IEEE
Published
2021-06
Pages
9842-9852
ISSN
1063-6919 (ISSN-L)
DOI
10.1109/cvpr46437.2021.00972
OpenAlex
W3171353004
Cited by
175
References
83
Citations by year
2026: 22 · 2025: 39 · 2024: 49 · 2023: 40 · 2022: 17 · 2021: 8
FWCI
9.32
Type
conference paper
Language
en
Open access
closed
Licence
Locations
IEEE Computer Society Conference on Computer Vision and Pattern Recognition/Proceedings - IEEE Computer Society Conference on Computer Vision and Pattern Recognition/Proceedings (publishedVersion)
Topics
  • Physical Sciences › Computer Science › Computer Vision and Pattern Recognition › Multimodal Machine Learning Applications
  • Physical Sciences › Computer Science › Computer Vision and Pattern Recognition › Visual Attention and Saliency Detection
  • Physical Sciences › Computer Science › Computer Vision and Pattern Recognition › Advanced Neural Network Applications
Keywords
causal attention, vision-language models, spurious correlations, confounding, front-door adjustment, causal intervention, LXMERT, pre-training, model generalization

About these checks

Registry, retraction, Cites and statistics results are computed, not judged, with one exception: a printed reference that no index lists is read once by a model before it counts as a citation error. Claims are extracted by a model and kept only when their quote appears verbatim in the text; each cited source is then checked, with a model’s help, for whether it says what the claim relies on. Those verdicts can be wrong — read the claim’s quote and the cited source before relying on one.

Statistics and arithmetic are recomputed, not judged.

Confidence bands are shown only for fields calibrated on our annotated test set. None is calibrated yet, so no band is shown.

What counts as a claim ›How we measure confidence ›How we check this ›

Are you an author, or do you see a problem with this page? Request a review