Pith. sign in

REVIEW 3 major objections 4 minor 34 references

Cause-Effect Driven Optimization for Robust Medical Visual Question Answering with Language Biases

T0 review · 3 major / 4 minor · reviewed 2026-08-06 · deepseek-v4-flash

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗ Review JSON

Pith's one-line read The paper claims that a training-time optimization framework, CEDO, can remove language biases in medical visual question answering by jointly correcting answer-category imbalance and question-modality shortcut, and reports…

desk verdict The empirical wrapper looks useful, but the GMS gradient equations are internally inconsistent, so the debiasing mechanism as written cannot explain the gains. read the letter →

arxiv 2506.17903 v1 pith:IJGVA7UP submitted 2025-06-22 cs.CV cs.AI

Huanjia Zhu , Yishu Liu , Xiaozhao Fang , Guangming Lu , Bingzhi Chen This is my paper · ORCID
classification cs.CVcs.AI
keywords medicalvisualquestionansweringlanguagebiasmitigationmultimodallearninggradientorthogonalityParetooptimizationlossrescalingout-of-distributiongeneralization
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The reading

The paper claims that language biases in medical visual question answering (Med-VQA) can be mitigated by a training-time optimization framework, CEDO, that treats the bias's cause (answer-category imbalance) and its effect (the question modality becoming a shortcut) separately. On two newly constructed out-of-distribution benchmarks, SLAKE-CP and VQA-RAD-CP, CEDO improves accuracy by 2.85 and 2.62 points over the state of the art, while also raising accuracy on the standard SLAKE and VQA-RAD datasets. The appeal is that a generic wrapper around an existing Med-VQA model can make it more robust to spurious question-answer correlations without losing in-distribution performance.

What carries the argument

The load-bearing mechanism is a three-part gradient and loss transformation. First, MHO partitions the model parameters into question, image, and classifier groups and updates each with its own learning rate, slowing the bias-prone question encoder. Second, GMS computes joint and unimodal losses, combines their gradients with adaptive Pareto weights, then replaces each gradient by itself minus its projection onto the others, an orthogonality constraint intended to delete the bias component from the question gradient. Third, DLR multiplies the supervised contrastive term by a weight $W_j^i = \log(1+\exp(1/(M_j m_j^i)))$ per sample, where $M_j$ is the number of samples of question type $j$ and $m_j^i$ the number of answers of category $i$ under that type, so rare answers are not starved. Together these modules are meant to prevent the question encoder from dominating and keep the loss balanced across answer categories.

What would settle it

On a batch of SLAKE-CP, compute the cosine similarity between $G_q$ and $G_v$ and the norm of $G_q$ before and after the orthogonality projection. If, averaged over batches, the cosine similarity is already near zero—or the projection barely changes $G_q$—then the orthogonality step has nothing to remove, and any accuracy gain must come from MHO or DLR rather than GMS.

Watch

Extended reading notes

Core claim

CEDO is built on the observation that a Med-VQA model's answers are distorted by two linked biases: imbalanced answer distributions cause the question type to correlate spuriously with frequent answers, and during training the question encoder receives oversized gradient updates that turn it into a shortcut. The framework attacks both sides: MHO assigns a lower learning rate to the question encoder and higher rates to the image encoder and classifier; GMS combines the joint and unimodal losses with a Pareto-optimized weighted gradient and then subtracts the projection of each gradient onto the others to enforce orthogonality; DLR reweights the supervised contrastive loss by an inverse question-type/answer frequency with a softplus smoothing. After integrating these into the base model, the paper reports state-of-the-art results on biased benchmarks (79.27 on SLAKE-CP, 92.07 on VQA-RAD-CP) and gains of 1.98 and 0.67 points on the standard SLAKE and VQA-RAD datasets.

Load-bearing premise

The whole mechanism hinges on the gradients used in the Pareto combination actually being defined for each modality's own parameters, and on the projection step removing only the bias component of the question gradient rather than useful signal.

Editorial extensions

If this is right

  • If CEDO is correct, a base Med-VQA model can be made more robust to language bias purely by changing its optimization, without altering its architecture or training data.
  • The new benchmarks SLAKE-CP and VQA-RAD-CP provide a repeatable protocol for measuring language-bias robustness in the medical domain, following the VQA-CP v2 ratio.
  • The three modules are additive: ablations show each alone improves SLAKE-CP (MHO +1.59, GMS +1.91, DLR +2.52) and the full combination +2.85 over baseline, suggesting the cause/effect split captures complementary failure modes.
  • The paper reports that CEDO also improves accuracy on the large-scale natural-domain VQA-CE benchmark, indicating the method is not limited to the two constructed datasets.

Reading between the lines

Editorial extensions of the paper, not claims the author makes directly.

  • The cause/effect decomposition suggests a general recipe for debiasing other multimodal systems: correct the data-level imbalance before it enters the loss, and constrain gradient updates to stop the dominant modality from absorbing the bias.
  • If the gradient-orthogonality step is really doing the work, the same projection could be applied to other bias-carrying gradient directions (e.g., location priors in chest X-ray diagnosis) as long as the bias direction can be identified.
  • One way to test the mechanism directly is to ablate the projection direction: replace it with a random unit vector of the same norm; if accuracy does not drop, the orthogonality constraint is not the active ingredient.
  • The reported parameter analysis keeps $\eta_q$ smaller than $\eta_v$ and $\eta_c$ (0.002 vs 0.003), an asymmetry consistent with the paper's story and a cheap sensitivity check: sweep the three rates independently and see if the advantage disappears when $\eta_q \ge \eta_v$.
Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, and a circularity audit.

Referee Report

3 major / 4 minor

Summary. This paper proposes CEDO, a training-time optimization wrapper for Med-VQA that targets language bias from what it calls cause and effect sides. The method combines three modules: MHO, which assigns separate learning rates to question, image, and classifier parameters; GMS, which uses a Pareto-style combination of unimodal and joint gradients together with a gradient-projection correction; and DLR, which reweights the supervised contrastive loss according to inverse question-type/answer co-occurrence counts. The authors construct two bias-sensitive benchmarks, SLAKE-CP and VQA-RAD-CP, by redistributing samples per question type, and report state-of-the-art accuracy on SLAKE, VQA-RAD, their CP variants, and VQA-CE, with the largest gains on the CP benchmarks. The central claim is that this optimization wrapper comprehensively mitigates language biases while preserving or slightly improving standard accuracy.

Significance. If the proposed mechanism worked as described, CEDO would be a practical contribution: it is an optimization-level wrapper that does not require architecture changes, counterfactual sample generation, or additional inference branches, and it reports consistent gains on both in-distribution and out-of-distribution Med-VQA benchmarks. The ablation study in Table 4 gives initial evidence that each module contributes, and the experiments span five datasets including the natural-domain VQA-CE. However, the current manuscript does not provide a faithful specification of the GMS mechanism in Eqs. (8)-(13), and the headline differences are not supported by error bars or a validation protocol that avoids tuning on the test benchmark. The benchmark construction also needs clearer differentiation from the DeBCF protocol described in [Zhan et al., 2023]. For these reasons, the empirical claim is promising but cannot yet be fully credited as stated.

major comments (3)
  1. [§3.3, Eqs. (8)-(13)] Eq. (8) defines G_k = ∇_{θ_q} L_k(R_k, A) for all k ∈ {t, q, v}, but the image unimodal loss L_v depends on θ_v and on the shared fusion and classifier parameters, not on θ_q; taken literally this makes G_v ≡ 0, so the Pareto combination in Eq. (9) does not balance modality gradients and the projection in Eq. (12) becomes degenerate. If the intended definition is a gradient over the full parameter set, that definition is never stated, and the projection in Eq. (11) is not a valid orthogonalization: subtracting proj_{G_q}(G_v) from G_q does not make G'_q orthogonal to G_v, and in the aligned case G_q = G_v it erases the entire agreed update rather than removing a biased component. Because the reported 2.85 and 2.62 point CP gains are attributed to this GMS mechanism, the paper currently does not specify a well-defined debiasing algorithm that matches its own causal narrative.
  2. [§4.3, Tables 1-3; §4.5, Fig. 4] The reported improvements are not accompanied by error bars or significance tests, and the hyperparameters η_q, η_v, and η_c are selected on SLAKE-CP itself (Fig. 4), which is the same benchmark used for the headline comparison in Table 1. This makes it difficult to assess whether the 0.67-point gain on SLAKE or the 1.23-point gain on VQA-CE is distinguishable from run-to-run variation, and whether part of the SLAKE-CP advantage comes from test-set tuning. Please report multiple-seed means and standard deviations and move hyperparameter selection to a validation split.
  3. [§4.2, Bias Reconstruction] The section says the CP datasets are built "following [Zhan et al., 2023]" and at the same time calls them "novel," so the relationship to the DeBCF bias-sensitive dataset should be clarified. In particular, the authors should state what is new relative to DeBCF, whether the same split protocol is reused, and how question types and "Other" answers are counted in the 39:1, 1:39, and 3:1 allocations. Without this information, the reader cannot judge whether the benchmarks are genuinely new or whether the evaluation is partly self-referential with respect to the dataset construction protocol.
minor comments (4)
  1. [§3.4, Eq. (16)] The positive set is defined as P_i = {p ∈ I s.t. a_p = a_j}, but j is not defined in the formula; it should presumably be a_i, the ground-truth answer of sample i.
  2. [Table 1, RUBi row] The RUBi row on VQA-RAD-CP reports All=81.27 with Open=60.62 and Closed=86.13; please clarify whether the All column is a weighted average over the open and closed subsets or a separately computed metric, since the three values are not mutually consistent without additional information.
  3. [Fig. 4 and §4.5] The axis labels of Fig. 4 appear to show η_q × 1000, η_v × 1000, and η_c × 1000; the caption should state explicitly that the displayed values are the true learning rates divided by 1000, and the text should indicate which hyperparameters are held fixed while each one is varied.
  4. [§4.2 and Datasets] The paper does not state whether the constructed SLAKE-CP and VQA-RAD-CP splits and the training code will be released; for a benchmark contribution this information is important for reproducibility and for comparisons by later work.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: the empirical claims are self-contained and no derivation reduces to its own inputs.

full rationale

The paper's central claim is an empirical comparison, not a derived prediction: CEDO is an optimization wrapper whose three mechanisms (MHO per-modality learning rates, GMS Pareto/orthogonality gradient combination, DLR inverse-frequency loss rescaling) are each defined from the model's own losses and data statistics, not from the target accuracy. Eq. (4) updates each parameter group with its own learning rate; Eqs. (5)-(9) define unimodal losses and a Pareto combination; Eq. (14) defines DLR weights from label counts. None of these equations is equivalent to the reported SLAKE-CP/VQA-RAD-CP accuracies, so there is no equation-level circularity. The hyperparameters eta_q, eta_v, eta_c are selected on SLAKE-CP (Fig. 4) and then reported on SLAKE-CP, which is a benchmark-tuning concern rather than a circular derivation: the result is not forced by construction. The SLAKE-CP and VQA-RAD-CP construction follows the external VQA-CP v2 protocol and Zhan et al.; the authors built the benchmarks, but the method does not assume the benchmark answers in defining its losses or gradients. The only self-citation (Zhu et al. 2024) appears as a baseline in Table 3 and is not load-bearing. The internal inconsistency in Eqs. (8) and (12) — G_v as written is a gradient with respect to theta_q and is therefore zero, and Eq. (12) does not orthogonalize G_q against G_v — is a serious correctness and specification defect, but it is not circularity: it does not make a prediction equal to its input, it makes the described mechanism ill-defined. Under the stated circularity criteria, this paper warrants score 0.

Assumptions & free parameters 4 free parameters · 6 assumptions · 0 invented entities

The framework contributes no new physical or architectural entities; its working parts are borrowed mechanisms. The load-bearing choices are three hand-selected learning rates, the CP split ratios, and several domain heuristics about where bias lives and how gradients convey it. The Pareto theorem is the only standard mathematical input, and it is used as a black box.

free parameters (4)
  • Question learning rate eta_q = 0.002
    Hand-searched in Section 4.5 on SLAKE-CP accuracy; MHO's ability to slow question-modality learning depends on this value.
  • Image learning rate eta_v = 0.003
    Hand-searched in Section 4.5; the claimed benefit of heterogeneous optimization depends on this relative rate.
  • Classifier learning rate eta_c = 0.003
    Hand-searched in Section 4.5; the separate learning rate for the classifier is introduced in Eq (4).
  • CP benchmark split ratios = 39:1, 1:39, 3:1
    Hand-chosen in Section 4.2 following VQA-CP v2; the difficulty of SLAKE-CP and VQA-RAD-CP, and therefore the size of the reported gains, is determined by these ratios.
assumptions (6)
  • standard math MGDA/Pareto theorem from Desideri (2012): minimizing the norm of a weighted sum of gradients yields a Pareto-stationary point or a descent direction for all objectives.
    Invoked in Section 3.3 to justify Eq (9); accepted mathematical result from cited literature.
  • domain assumption The question modality is intrinsically more biased and the image modality needs a higher learning rate.
    Motivates MHO in Section 3.2; based on citations rather than on measurements in this paper.
  • domain assumption Data imbalance is a primary cause of language bias and is transmitted into shortcut bias through gradients.
    The causal story in Section 1 and Figure 1; asserted without a formal causal model.
  • ad hoc to paper Subtracting the projection of one modality gradient onto another removes only the biased component.
    Eq (12) in Section 3.3 assumes the removed projection is exactly the shortcut; no proof or experiment isolates this component.
  • ad hoc to paper The inverse co-occurrence weight w_j^i = 1/(M_j m_j^i), smoothed by softplus, is the appropriate rescaling for answer imbalance.
    Eqs (14)-(15) in Section 3.4; chosen heuristically with no alternative weighting compared.
  • domain assumption Question type defined by initial words is the correct grouping for both bias construction and loss weighting.
    Used in Section 4.2 to build CP splits and in Section 3.4 to compute weights; assumes the bias is aligned with the first words of the question.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Cause-Effect Driven Optimization for Robust Medical Visual Question Answering with Language Biases." pith.science (2026). https://pith.science/paper/IJGVA7UP

@misc{pith2026250617903,
  author       = {Pith},
  title        = {Pith review of: Cause-Effect Driven Optimization for Robust Medical Visual Question Answering with Language Biases},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/IJGVA7UP}},
  note         = {Machine review of arXiv:2506.17903}
}
read the original abstract

Existing Medical Visual Question Answering (Med-VQA) models often suffer from language biases, where spurious correlations between question types and answer categories are inadvertently established. To address these issues, we propose a novel Cause-Effect Driven Optimization framework called CEDO, that incorporates three well-established mechanisms, i.e., Modality-driven Heterogeneous Optimization (MHO), Gradient-guided Modality Synergy (GMS), and Distribution-adapted Loss Rescaling (DLR), for comprehensively mitigating language biases from both causal and effectual perspectives. Specifically, MHO employs adaptive learning rates for specific modalities to achieve heterogeneous optimization, thus enhancing robust reasoning capabilities. Additionally, GMS leverages the Pareto optimization method to foster synergistic interactions between modalities and enforce gradient orthogonality to eliminate bias updates, thereby mitigating language biases from the effect side, i.e., shortcut bias. Furthermore, DLR is designed to assign adaptive weights to individual losses to ensure balanced learning across all answer categories, effectively alleviating language biases from the cause side, i.e., imbalance biases within datasets. Extensive experiments on multiple traditional and bias-sensitive benchmarks consistently demonstrate the robustness of CEDO over state-of-the-art competitors.

Figures

Figures reproduced from arXiv: 2506.17903 by the authors.

Figure 1
Figure 1. The causal mechanism underlying language biases in [PITH_FULL_IMAGE:figures/full_fig_p001_1.png] view at source ↗
Figure 2
Figure 2. Illustration of the proposed Cause-Effect Driven Optimization (CEDO) framework for addressing medical language biases. [PITH_FULL_IMAGE:figures/full_fig_p003_2.png] view at source ↗
Figure 3
Figure 3. (a) Designed distribution bias of training set and testing [PITH_FULL_IMAGE:figures/full_fig_p005_3.png] view at source ↗
Figures from the paper (2 more)
Figure 4
Figure 4. Figure 4: Comparison of Accuracy on the SLAKE-CP dataset with [PITH_FULL_IMAGE:figures/full_fig_p007_4.png]
Figure 5
Figure 5. Figure 5: Visualization analysis of CEDO. Our proposed approach [PITH_FULL_IMAGE:figures/full_fig_p007_5.png]

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

34 extracted references · 33 canonical work pages

  1. [1]

    Don’t just assume; look and answer: Overcoming priors for visual question answering

    [Agrawalet al., 2018 ] Aishwarya Agrawal, Dhruv Batra, Devi Parikh, and Aniruddha Kembhavi. Don’t just assume; look and answer: Overcoming priors for visual question answering. InProceedings of the IEEE Conference on Computer Vision and Pattern Recognition (CVPR), pages 4971–4980,

    work page 2018

  2. [6]

    Multi-modal masked autoencoders for medical vision-and- language pre-training

    [Chenet al., 2022 ] Zhihong Chen, Yuhao Du, Jinpeng Hu, Yang Liu, Guanbin Li, Xiang Wan, and Tsung-Hui Chang. Multi-modal masked autoencoders for medical vision-and- language pre-training. InMedical Image Computing and Computer Assisted Intervention (MICCAI), pages 679–

    work page 2022

  3. [8]

    Beyond question- based biases: Assessing multimodal shortcut learning in visual question answering

    [Dancetteet al., 2021 ] Corentin Dancette, Remi Cadene, Damien Teney, and Matthieu Cord. Beyond question- based biases: Assessing multimodal shortcut learning in visual question answering. InProceedings of the IEEE In- ternational Conference on Computer Vision (ICCV), pages 1574–1583,

    work page 2021

  4. [11]

    Loss re-scaling vqa: Re- visiting the language prior problem from a class-imbalance view.IEEE Transactions on Image Processing (TIP), 31:227–238,

    [Guoet al., 2021 ] Yangyang Guo, Liqiang Nie, Zhiyong Cheng, Qi Tian, and Min Zhang. Loss re-scaling vqa: Re- visiting the language prior problem from a class-imbalance view.IEEE Transactions on Image Processing (TIP), 31:227–238,

    work page 2021

  5. [12]

    Greedy gradient ensemble for robust visual question an- swering

    [Hanet al., 2021 ] Xinzhe Han, Shuhui Wang, and Chi Su. Greedy gradient ensemble for robust visual question an- swering. InProceedings of the IEEE International Confer- ence on Computer Vision (ICCV), pages 1584–1593,

    work page 2021

  6. [13]

    General greedy de-bias learn- ing.IEEE Transactions on Pattern Analysis and Machine Intelligence (TPAMI), 45:1–17,

    [Hanet al., 2023 ] Xinzhe Han, Shuhui Wang, Chi Su, Qing- ming Huang, and Qi Tian. General greedy de-bias learn- ing.IEEE Transactions on Pattern Analysis and Machine Intelligence (TPAMI), 45:1–17,

    work page 2023

  7. [14]

    Deep learning techniques for medical image segmentation: achievements and challenges.Journal of Digital Imaging (JDI), 32:582–596,

    [Hesamianet al., 2019 ] Mohammad Hesam Hesamian, Wenjing Jia, Xiangjian He, and Paul Kennedy. Deep learning techniques for medical image segmentation: achievements and challenges.Journal of Digital Imaging (JDI), 32:582–596,

    work page 2019

  8. [15]

    Modality competi- tion: What makes joint training of multi-modal network fail in deep learning?(provably)

    [Huanget al., 2022 ] Yu Huang, Junyang Lin, Chang Zhou, Hongxia Yang, and Longbo Huang. Modality competi- tion: What makes joint training of multi-modal network fail in deep learning?(provably). InProceedings of the International Conference on Machine Learning (ICML), pages 9226–9259. PMLR,

    work page 2022

Show all 34 references
  1. [16]

    Leveraging image complexity in macro-level neural network design for medical image segmentation

    [Khanet al., 2022 ] Tariq M Khan, Syed S Naqvi, and Erik Meijering. Leveraging image complexity in macro-level neural network design for medical image segmentation. Scientific Reports, 12(1):22286,

    2022

  2. [17]

    Supervised contrastive learning

    [Khoslaet al., 2020 ] Prannay Khosla, Piotr Teterwak, Chen Wang, Aaron Sarna, Yonglong Tian, Phillip Isola, Aaron Maschinot, Ce Liu, and Dilip Krishnan. Supervised contrastive learning. InProceedings of the Conference and Workshop on Neural Information Processing Systems (Neur...

    2020

  3. [19]

    A dataset of clinically generated visual questions and answers about ra- diology images.Scientific data, 5(1):1–10,

    [Lauet al., 2018 ] Jason J Lau, Soumya Gayen, Asma Ben Abacha, and Dina Demner-Fushman. A dataset of clinically generated visual questions and answers about ra- diology images.Scientific data, 5(1):1–10,

    2018

  4. [20]

    Lpf: A language-prior feedback objective function for de-biased visual question answering

    [Lianget al., 2021 ] Zujie Liang, Haifeng Hu, and Jiaying Zhu. Lpf: A language-prior feedback objective function for de-biased visual question answering. InProceedings of the 44th International ACM SIGIR Conference on Re- search and Development in Information Retrieval (SIGIR)...

    2021

  5. [21]

    Medical visual question answering via condi- tional reasoning and contrastive learning.IEEE Transac- tions on Medical Imaging (TMI), 42(5):1532–1545,

    [Liuet al., 2022 ] Bo Liu, Li-Ming Zhan, Li Xu, and Xiao- Ming Wu. Medical visual question answering via condi- tional reasoning and contrastive learning.IEEE Transac- tions on Medical Imaging (TMI), 42(5):1532–1545,

    2022

  6. [22]

    Overcoming data limitation in medical visual ques- tion answering

    [Nguyenet al., 2019 ] Binh D Nguyen, Thanh-Toan Do, Binh X Nguyen, Tuong Do, Erman Tjiputra, and Quang D Tran. Overcoming data limitation in medical visual ques- tion answering. InMedical Image Computing and Com- puter Assisted Intervention (MICCAI), pages 522–530. Springer,

    2019

  7. [23]

    Balanced multimodal learning via on-the-fly gradient modulation

    [Penget al., 2022 ] Xiaokang Peng, Yake Wei, Andong Deng, Dong Wang, and Di Hu. Balanced multimodal learning via on-the-fly gradient modulation. InProceed- ings of the IEEE Conference on Computer Vision and Pat- tern Recognition (CVPR), pages 8238–8247,

    2022

  8. [24]

    Overcoming lan- guage priors in visual question answering with adversarial regularization

    [Ramakrishnanet al., 2018 ] Sainandan Ramakrishnan, Aishwarya Agrawal, and Stefan Lee. Overcoming lan- guage priors in visual question answering with adversarial regularization. InProceedings of the Conference and Workshop on Neural Information Processing Systems (NeurIPS), volume 31,

    2018

  9. [25]

    Focal loss for dense object detection

    [Ross and Doll´ar, 2017] T-YLPG Ross and GKHP Doll ´ar. Focal loss for dense object detection. InProceedings of the IEEE Conference on Computer Vision and Pattern Recog- nition (CVPR), pages 2980–2988,

    2017

  10. [27]

    Embracing imperfect datasets: A review of deep learning solutions for medical image segmentation.Medical Image Analysis (MedIA), 63:101693,

    [Tajbakhshet al., 2020 ] Nima Tajbakhsh, Laura Jeyaseelan, Qian Li, Jeffrey N Chiang, Zhihao Wu, and Xiaowei Ding. Embracing imperfect datasets: A review of deep learning solutions for medical image segmentation.Medical Image Analysis (MedIA), 63:101693,

    2020

  11. [28]

    A question-centric model for visual question answering in medical imaging.IEEE Transactions on Medical Imaging (TMI), 39(9):2856– 2868,

    [Vuet al., 2020 ] Minh H Vu, Tommy L ¨ofstedt, Tufve Ny- holm, and Raphael Sznitman. A question-centric model for visual question answering in medical imaging.IEEE Transactions on Medical Imaging (TMI), 39(9):2856– 2868,

    2020

  12. [29]

    What makes training multi-modal classification net- works hard? InProceedings of the IEEE Conference on Computer Vision and Pattern Recognition (CVPR), pages 12695–12705,

    [Wanget al., 2020 ] Weiyao Wang, Du Tran, and Matt Feis- zli. What makes training multi-modal classification net- works hard? InProceedings of the IEEE Conference on Computer Vision and Pattern Recognition (CVPR), pages 12695–12705,

    2020

  13. [30]

    Gradient-guided modality decoupling for missing-modality robustness

    [Wanget al., 2024 ] Hao Wang, Shengda Luo, Guosheng Hu, and Jianguo Zhang. Gradient-guided modality decoupling for missing-modality robustness. InProceedings of the AAAI Conference on Artificial Intelligence (AAAI), vol- ume 38, pages 15483–15491,

    2024

  14. [33]

    Debias- ing medical visual question answering via counterfactual training

    [Zhanet al., 2023 ] Chenlu Zhan, Peng Peng, Hanrong Zhang, Haiyue Sun, Chunnan Shang, Tao Chen, Hongsen Wang, Gaoang Wang, and Hongwei Wang. Debias- ing medical visual question answering via counterfactual training. InMedical Image Computing and Computer As- sisted Interventio...

    2023

  15. [34]

    Combating visual question answering hallucinations via robust multi-space co-debias learning

    [Zhuet al., 2024 ] Jiawei Zhu, Yishu Liu, Huanjia Zhu, Hui Lin, Yuncheng Jiang, Zheng Zhang, and Bingzhi Chen. Combating visual question answering hallucinations via robust multi-space co-debias learning. InProceedings of the 32nd ACM International Conference on Multimedia, pa...

    2024

  16. [2012]

    Mul- tiple meta-model quantifying for medical visual question answering

    [Doet al., 2021 ] Tuong Do, Binh X Nguyen, Erman Tjipu- tra, Minh Tran, Quang D Tran, and Anh Nguyen. Mul- tiple meta-model quantifying for medical visual question answering. InMedical Image Computing and Computer Assisted Intervention (MICCAI), pages 64–74. Springer,

    2021

  17. [2015]

    Venkatesh Babu

    [Basuet al., 2023 ] Abhipsa Basu, Sravanti Addepalli, and R. Venkatesh Babu. Rmlvqa: A margin loss approach for visual question answering with language biases. InPro- ceedings of the IEEE Conference on Computer Vision and Pattern Recognition (CVPR), pages 11671–11680,

    2023

  18. [2016]

    Multi-modal factorized bilinear pooling with co-attention learning for visual question answering

    [Yuet al., 2017 ] Zhou Yu, Jun Yu, Jianping Fan, and Dacheng Tao. Multi-modal factorized bilinear pooling with co-attention learning for visual question answering. InProceedings of the IEEE Conference on Computer Vi- sion and Pattern Recognition (CVPR), pages 1821–1830,

    2017

  19. [2017]

    Multi-task learning as multi-objective optimization

    [Sener and Koltun, 2018] Ozan Sener and Vladlen Koltun. Multi-task learning as multi-objective optimization. In Proceedings of the Conference and Workshop on Neural Information Processing Systems (NeurIPS), volume 31,

    2018

  20. [2018]

    Vqa: Visual question answering

    [Antolet al., 2015 ] Stanislaw Antol, Aishwarya Agrawal, Ji- asen Lu, Margaret Mitchell, Dhruv Batra, C Lawrence Zit- nick, and Devi Parikh. Vqa: Visual question answering. InProceedings of the IEEE International Conference on Computer Vision (ICCV), pages 2425–2433,

    2015

  21. [2019]

    Rubi: Re- ducing unimodal biases in visual question answering

    [Cadeneet al., 2019 ] Remi Cadene, Corentin Dancette, Hedi Ben-younes, Matthieu Cord, and Devi Parikh. Rubi: Re- ducing unimodal biases in visual question answering. In Proceedings of the Conference and Workshop on Neural Information Processing Systems (NeurIPS), volume 32,

    2019

  22. [2020]

    Bilinear attention networks

    [Kimet al., 2018 ] Jin-Hwa Kim, Jaehyun Jun, and Byoung- Tak Zhang. Bilinear attention networks. InProceedings of the Conference and Workshop on Neural Information Processing Systems (NeurIPS), volume 31,

    2018

  23. [2021]

    Multiple-gradient descent algorithm (mgda) for multiobjective optimization

    [D´esid´eri, 2012] Jean-Antoine D ´esid´eri. Multiple-gradient descent algorithm (mgda) for multiobjective optimization. Comptes Rendus Mathematique, 350(5-6):313–318,

    2012

  24. [2022]

    Don’t take the easy way out: Ensem- ble based methods for avoiding known dataset biases

    [Clarket al., 2019 ] Christopher Clark, Mark Yatskar, and Luke Zettlemoyer. Don’t take the easy way out: Ensem- ble based methods for avoiding known dataset biases. In Conference on Empirical Methods in Natural Language Processing (EMNLP),

    2019

  25. [2023]

    Vqa-med: Overview of the medical visual question answering task at imageclef

    [Ben Abachaet al., 2019 ] Asma Ben Abacha, Sadid A Hasan, Vivek V Datla, Dina Demner-Fushman, and Hen- ning M ¨uller. Vqa-med: Overview of the medical visual question answering task at imageclef

    2019

  26. [2024]

    Stacked attention net- works for image question answering

    [Yanget al., 2016 ] Zichao Yang, Xiaodong He, Jianfeng Gao, Li Deng, and Alex Smola. Stacked attention net- works for image question answering. InProceedings of the IEEE Conference on Computer Vision and Pattern Recog- nition (CVPR), pages 21–29,

    2016

Pith tools

Reviewed August 6, 2026 · model on record in the stance chip above.