Pith. sign in

REVIEW 4 major objections 5 minor 42 references

Bayesian Principles Improve Prompt Learning In Vision-Language Models

T0 review · 4 major / 5 minor · reviewed 2026-08-16 · deepseek-v4-flash

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

Pith's one-line read Treating classifier logits as Bayesian random variables, with one-vs-each softmax and Pólya-Gamma augmentation, improves prompt-learning generalization in CLIP.

desk verdict The empirical OVE-PG recipe shows consistent gains on unseen classes, but the Bayesian derivation does not survive contact with the A≈I approximation, and the paper's own appendix admits the KL term is an L2 penalty. read the letter →

arxiv 2504.14123 v1 pith:GU6O5DU6 submitted 2025-04-19 cs.AI cs.CLcs.CV

Mingyu Kim , Jongwoo Ko , Mijung Park This is my paper · ORCID
classification cs.AIcs.CLcs.CV
keywords promptlearningvision-languagemodelsCLIPPólya-Gammaaugmentationone-vs-eachsoftmaxBayesianinferencegeneralizationoverfitting
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 prompt learning in vision-language models can be made to generalize better by treating the classifier logits as Bayesian random variables rather than fixed outputs. Specifically, it replaces the standard softmax likelihood with a one-vs-each approximation—each class probability is built from sigmoids over pairwise logit differences—and augments those sigmoids with Pólya-Gamma latent variables. The fine-tuned model's logits are pulled toward the pretrained model's logits through a KL-divergence penalty, which acts as a knowledge-distillation regularizer. On unseen classes, this raises CoOp's average accuracy from 71.05 to 75.04, with particularly large gains on EuroSAT and FGVC Aircraft, while keeping the number of trainable parameters unchanged. The paper concludes that the combination is a simpler and more effective anti-overfitting device than extra regularization terms or extra networks.

What carries the argument

The central mechanism is a closed-form posterior sampler for logits built on the one-vs-each softmax bound and Pólya-Gamma augmentation. The pairwise-difference matrix $A \in \mathbb{R}^{C\times C\times C}$ maps logits $f$ to comparisons $\psi = Af$, with entries $f_{ni} - f_{nj}$; each sigmoid $\sigma(\psi_{nij})$ in the one-vs-each likelihood is augmented by a Pólya-Gamma variable $\omega$, which makes the conditional likelihood Gaussian in $\psi$. The paper approximates $A \approx I$ so the posterior covariance becomes diagonal, $(\alpha I + \Omega)^{-1}$, and samples fine-tuned logits from $\mathcal{N}(\mu_\theta, (\alpha I + \Omega)^{-1})$ while drawing $\omega$ from $\mathrm{PG}(1, \psi)$ with $\psi$ built from the pretrained mean. These samples feed a negative log-likelihood loss, and the KL term $\|\mu_\theta - \mu\|_2^2$ anchors the fine-tuned logits to the pretrained logits. The diagonal approximation is what makes the method computationally cheap, and it is also the step that carries the claim's Bayesian interpretation.

What would settle it

Re-run the CoOp experiments with the exact posterior covariance $(\alpha I + A^\top \Omega A)^{-1}$ in place of the diagonal approximation $(\alpha I + \Omega)^{-1}$: if unseen-class accuracy is unchanged, the diagonal approximation is not producing the gain; if training becomes unstable or accuracy drops, the method's stated posterior is not what it samples from.

Watch

Extended reading notes

Core claim

The paper's central claim is that a Bayesian treatment of the logit function itself—rather than of the prompt parameters—reduces overfitting in prompt learning for vision-language models. The paper puts a Gaussian prior on the logits whose mean is the pretrained model's logit output $\mu = I(x)^\top T(p^c)$, and models the fine-tuned logits as a Gaussian centered at $\mu_\theta = I(x)^\top T(p^c_\theta + r^c_\theta(x))$. It replaces the softmax likelihood with the one-vs-each approximation, a product of sigmoids over pairwise logit differences, and adds Pólya-Gamma auxiliary variables so the conditional likelihood is Gaussian and posterior sampling is closed form. A KL-divergence penalty $\beta\|\mu_\theta - \mu\|_2^2$ keeps the fine-tuned logits close to the pretrained ones. The combination, OVE-PG, is claimed to improve unseen-class accuracy and cross-dataset transfer across CoOp, CoCoOp, MaPLe, and APEX while adding no extra parameters or separate regularization terms.

Load-bearing premise

The load-bearing assumption is that the matrix $A$ that forms every pairwise class comparison can be treated as the identity, so the sampled logits are uncorrelated and the posterior covariance is diagonal; if this approximation is unreliable, the sampling procedure no longer draws from the stated Bayesian posterior.

Editorial extensions

If this is right

  • Unseen-class accuracy improves for every prompt-learning method tested, with CoOp rising from 71.05 to 75.04 on average and EuroSAT jumping from 55.20 to 71.23.
  • Cross-dataset transfer improves on average for both CoOp and CoCoOp, with the largest target gains on EuroSAT and DTD.
  • The Pólya-Gamma augmentation, not the OVE approximation by itself, is what produces the generalization gain; plain OVE can even hurt on the hardest datasets.
  • The KL-anchoring strength $\beta$ is robust across 0.2–0.5, so the method does not depend on a finely tuned hyperparameter.
  • Because the change is confined to the likelihood and sampling step, it can be added to existing prompt-learning methods without extra parameters or architectural changes.

Reading between the lines

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

  • This recipe could generalize beyond prompt tokens: any fine-tuning scheme that puts a pretrained linear classifier or logit head on a frozen encoder could anchor its logits to the pretrained output and use OVE-PG sampling, separating the benefit of the Bayesian logit prior from prompt parameterization.
  • The large gain on EuroSAT, a domain far from the pretraining distribution, suggests the regularizer may be most valuable under distribution shift; a controlled sweep over shift magnitude would test whether the advantage scales with domain distance.
  • The posterior logits are softened by Pólya-Gamma noise, so OVE-PG may also improve calibration or out-of-distribution detection on unseen classes; the paper does not report those metrics.
Share X Bluesky LinkedIn Reddit HN

Signed reviews

No signed human review yet.

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

4 major / 5 minor

Summary. The paper proposes OVE-PG, a prompt-learning method for vision-language models that replaces the standard softmax likelihood with a one-vs-each (OVE) softmax approximation and uses Pólya-Gamma augmentation to make the resulting sigmoid factors conjugate with Gaussian priors over logits. The prior mean is set to the logits of the frozen pretrained CLIP model, and the posterior mean to the logits of the trainable prompt model; a KL-type term between prior and posterior is added as a regularizer. The method is evaluated on seen-to-unseen generalization across ten datasets and on cross-dataset transfer, integrated into CoOp, CoCoOp, MaPLe, and APEX. The reported results show consistent improvements over the corresponding softmax baselines, with the largest gains on EuroSAT and FGVC Aircraft for CoOp.

Significance. If the empirical results hold, the contribution is practically useful: it is a simple plug-in modification that improves unseen-class and cross-dataset accuracy across several prompt-learning frameworks without adding network parameters, and the code is publicly available. The reported improvements are large on some datasets (e.g., EuroSAT from 55.20 to 71.23 for CoOp) and consistent across baselines and seeds, with standard deviations reported in the appendix. However, the paper's central theoretical claim, that the objective is derived from Bayesian principles, is not currently established: the derivation relies on an acknowledged diagonal-covariance approximation and on collapsing a KL divergence to an L2 term, and several admitted deviations place the method closer to a heuristic regularized logit-matching procedure than to a posterior-based method. The significance of the paper therefore depends on whether the authors can either repair the derivation or honestly reframe the contribution as an empirical regularizer with PG noise.

major comments (4)
  1. [Section 3.4, Eq. (17) and Appendix C.4.3]
  2. [Section 3.4 and Appendix C.3, Eq. (C.44)]
  3. [Algorithm 1, line 'fθ^(m) ← ∑_C ψθ^(m)']
  4. [Appendix C.4.2]
minor comments (5)
  1. [Section 1]
  2. [Section 5.1]
  3. [References]
  4. [Appendix F, Table F.6]
  5. [Appendix D]

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: OVE-PG's derivation and held-out evaluation are self-contained; the main caveats are approximations, not input-output equivalences.

full rationale

The central derivation is a variational objective: a Polya-Gamma augmented one-vs-each likelihood, a Gaussian prior centered at the frozen CLIP logits, and a Gaussian variational posterior whose mean is the tunable prompt logits. The KL term is reduced to ||μθ−μ||^2_2 by dropping terms that do not depend on the prompt parameters (Section 3.4, Eq. 17), and Appendix C.4.3 explicitly shows the rigorous KL expression before the reduction. This is an approximation, not a relabeling of a fitted quantity as a prediction. The benchmark claims are evaluated on held-out unseen classes and cross-dataset transfer (Tables 1 and 2), so the empirical result is not forced by the loss definition. The A≈I diagonalization in Section 3.4 and Appendix C.3 is a genuine modeling assumption, acknowledged in C.4.1 as 'motivated by purely computational reasons'; it affects whether the sampler matches the stated posterior, but it does not make the derivation circular. Appendix C.4.2 candidly notes that the construction 'deviates from the usual Bayesian construction,' which is a stated limitation of the Bayesian framing rather than evidence that the prediction reduces to its inputs. The only self-citation is Yang et al. 2024 (APEX, co-authored by Jongwoo Ko), used as a baseline and integration target; it is not load-bearing for the core derivation. No circular step meets the quoted-reduction bar.

Assumptions & free parameters 3 free parameters · 5 assumptions · 0 invented entities

The method depends on tuned constants β and α plus several acknowledged simplifications: the diagonal A ≈ I approximation, the one-shot PG sampler, and the reduction of the KL term to an L2 penalty. These are modeling choices rather than new entities; there are no invented physical or computational entities beyond the standard auxiliary PG variables.

free parameters (3)
  • β (KL weight) = not stated for main tables (sensitivity over 0.1-0.7, robust 0.2-0.5)
    Controls the strength of the logit-matching penalty L_KLD = β||μθ - μ||^2. The paper does not report the exact value used in Tables 1-3 or the selection rule, only a sensitivity analysis in Figure 3 and Table F.6.
  • α (prior precision) = not stated for CLIP experiments (synthetic uses 1 and 100)
    Inverse variance of the Gaussian prior on logits and a scale for sampling noise. The appendix says the setup follows Derakhshani et al. 2023, but the actual value used in the main experiments is not given.
  • M (number of Gibbs chains) = not stated
    Number of parallel posterior samples in Algorithm 1. The experiments do not report M; a larger M changes sampling variance and the effective regularization, so the value is needed for replication.
assumptions (5)
  • standard math Polya-Gamma augmentation identity (Eq. 1) expresses logistic likelihood terms as Gaussian mixtures over PG variables.
    Used to derive Gaussian likelihood and posterior updates; the proof is cited to Polson et al. 2013 and reproduced in Appendix B.
  • standard math One-vs-each product lower-bounds the softmax (Eq. 9 and Appendix A).
    Used to replace softmax with a product of sigmoids; proof given in Appendix A using the inequality (1 + sum α_k) ≤ product (1 + α_k).
  • domain assumption Frozen CLIP logits are a sensible prior mean for downstream tasks.
    Equation 7 sets the prior mean to pretrained CLIP logits. This assumption is plausible but not proven, and it is central to the KL term.
  • ad hoc to paper A ≈ I and component-wise independence of logits.
    Section 3.4 states 'we approximate A ≈ I.' This makes the posterior diagonal and computationally feasible, but it contradicts the variance-doubling note in Appendix C.3.
  • ad hoc to paper PG variable ω can be sampled once from the pretrained logits and held fixed during prompt optimization.
    Appendix C.4.2 acknowledges that this 'deviates from the usual Bayesian construction.' It makes the sampler cheap but breaks full Gibbs-style posterior inference.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Bayesian Principles Improve Prompt Learning In Vision-Language Models." pith.science (2026). https://pith.science/paper/GU6O5DU6

@misc{pith2026250414123,
  author       = {Pith},
  title        = {Pith review of: Bayesian Principles Improve Prompt Learning In Vision-Language Models},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/GU6O5DU6}},
  note         = {Machine review of arXiv:2504.14123}
}
read the original abstract

Prompt learning is a popular fine-tuning method for vision-language models due to its efficiency. It requires a small number of additional learnable parameters while significantly enhancing performance on target tasks. However, most existing methods suffer from overfitting to fine-tuning data, yielding poor generalizability. To address this, we propose a new training objective function based on a Bayesian learning principle to balance adaptability and generalizability. We derive a prior over the logits, where the mean function is parameterized by the pre-trained model, while the posterior corresponds to the fine-tuned model. This objective establishes a balance by allowing the fine-tuned model to adapt to downstream tasks while remaining close to the pre-trained model.

Figures

Figures reproduced from arXiv: 2504.14123 by the authors.

Figure 1
Figure 1. 1D synthetic classification comparison between SoftMax, SoftMax_OVE and SoftMAX_OVE_PG [PITH_FULL_IMAGE:figures/full_fig_p005_1.png] view at source ↗
Figure 2
Figure 2. Effect of OVE-PG on ViT using the EMNIST and MNIST datasets. Our proposed OVE-PG ( [PITH_FULL_IMAGE:figures/full_fig_p006_2.png] view at source ↗
Figure 3
Figure 3. Sensitivity analysis of the hyperparameter [PITH_FULL_IMAGE:figures/full_fig_p009_3.png] view at source ↗

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

42 extracted references · 32 canonical work pages

  1. [1]

    Bossard, L., Guillaumin, M., and Gool, L. V. (2014a). Food-101 - mining discriminative components with random forests. In European Conference on Computer Vision

    work page 2014

  2. [2]

    Bossard, L., Guillaumin, M., and Gool, L. V. (2014b). Food-101--mining discriminative components with random forests. In European Conference on Computer Vision

    work page 2014

  3. [3]

    Chen, G., Yao, W., Song, X., Li, X., Rao, Y., and Zhang, K. (2023). PLOT : Prompt learning with optimal transport for vision-language models. In The Eleventh International Conference on Learning Representations

    work page 2023

  4. [4]

    D., Joo, W., and Moon, I.-C

    Cho, Y., Bae, H., Shin, S., Youn, Y. D., Joo, W., and Moon, I.-C. (2024). Make prompts adaptable: Bayesian modeling for vision-language prompt learning with data-dependent prior. In Proceedings of the AAAI Conference on Artificial Intelligence , volume 38, pages 11552--11560

    work page 2024

  5. [5]

    Cimpoi, M., Maji, S., Kokkinos, I., Mohamed, S., and Vedaldi, A. (2014). Describing textures in the wild. In IEEE Conference on Computer Vision and Pattern Recognition

    work page 2014

  6. [6]

    Deng, J., Dong, W., Socher, R., Li, L.-J., Li, K., and Fei-Fei, L. (2009). ImageNet : A large-scale hierarchical image database. In IEEE Conference on Computer Vision and Pattern Recognition

    work page 2009

  7. [7]

    M., Sanchez, E., Bulat, A., da Costa, V

    Derakhshani, M. M., Sanchez, E., Bulat, A., da Costa, V. G. T., Snoek, C. G., Tzimiropoulos, G., and Martinez, B. (2023a). Bayesian prompt learning for image-language model generalization. In Proceedings of the IEEE/CVF International Conference on Computer Vision , pages 15237--15246

    work page 2023

  8. [8]

    M., Sanchez, E., Bulat, A., da Costa, V

    Derakhshani, M. M., Sanchez, E., Bulat, A., da Costa, V. G. T., Snoek, C. G., Tzimiropoulos, G., and Martinez, B. (2023b). Bayesian prompt learning for image-language model generalization. In Proceedings of the IEEE/CVF International Conference on Computer Vision (ICCV) , pages 15237--15246

    work page 2023

Show all 42 references
  1. [9]

    Dosovitskiy, A., Beyer, L., Kolesnikov, A., Weissenborn, D., Zhai, X., Unterthiner, T., Dehghani, M., Minderer, M., Heigold, G., Gelly, S., Uszkoreit, J., and Houlsby, N. (2021). An image is worth 16x16 words: Transformers for image recognition at scale. In International Confe...

    2021

  2. [10]

    Fei-Fei, L., Fergus, R., and Perona, P. (2004a). Learning generative visual models from few training examples: An incremental bayesian approach tested on 101 object categories. 2004 Conference on Computer Vision and Pattern Recognition Workshop , pages 178--178

    2004

  3. [11]

    Fei-Fei, L., Fergus, R., and Perona, P. (2004b). Learning generative visual models from few training examples: An incremental bayesian approach tested on 101 object categories. In IEEE Conference on Computer Vision and Pattern Recognition - Workshops

    2004

  4. [12]

    Helber, P., Bischke, B., Dengel, A., and Borth, D. (2019). Eurosat: A novel dataset and deep learning benchmark for land use and land cover classification. IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing

    2019

  5. [13]

    Jia, C., Yang, Y., Xia, Y., Chen, Y.-T., Parekh, Z., Pham, H., Le, Q., Sung, Y.-H., Li, Z., and Duerig, T. (2021). Scaling up visual and vision-language representation learning with noisy text supervision. In International conference on machine learning , pages 4904--4916. PMLR

    2021

  6. [14]

    U., Rasheed, H., Maaz, M., Khan, S., and Khan, F

    Khattak, M. U., Rasheed, H., Maaz, M., Khan, S., and Khan, F. S. (2023a). Maple: Multi-modal prompt learning. In The IEEE/CVF Conference on Computer Vision and Pattern Recognition

    2023

  7. [15]

    U., Wasim, S

    Khattak, M. U., Wasim, S. T., Naseer, M., Khan, S., Yang, M.-H., and Khan, F. S. (2023b). Self-regulating prompts: Foundational model adaptation without forgetting. In Proceedings of the IEEE/CVF International Conference on Computer Vision , pages 15190--15200

    2023

  8. [16]

    Krause, J., Stark, M., Deng, J., and Fei-Fei, L. (2013). 3D object representations for fine-grained categorization. In IEEE Conference on Computer Vision and Pattern Recognition - Workshops

    2013

  9. [17]

    Lester, B., Al-Rfou, R., and Constant, N. (2021). The power of scale for parameter-efficient prompt tuning. In Conference on Empirical Methods in Natural Language Processing

    2021

  10. [18]

    J., and Adams, R

    Linderman, S., Johnson, M. J., and Adams, R. P. (2015). Dependent multinomial models made easy: Stick-breaking with the polya-gamma augmentation. In Cortes, C., Lawrence, N., Lee, D., Sugiyama, M., and Garnett, R., editors, Advances in Neural Information Processing Systems , v...

    2015

  11. [19]

    Liu, X., Zheng, Y., Du, Z., Ding, M., Qian, Y., Yang, Z., and Tang, J. (2023). Gpt understands, too. AI Open

    2023

  12. [20]

    Lu, Y., Liu, J., Zhang, Y., Liu, Y., and Tian, X. (2022). Prompt distribution learning. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , pages 5206--5215

    2022

  13. [21]

    Maji, S., Rahtu, E., Kannala, J., Blaschko, M., and Vedaldi, A. (2013). Fine-grained visual classification of aircraft. arXiv preprint arXiv:1306.5151

    2013 arXiv

  14. [22]

    Polson, J

    Nicholas G. Polson, J. G. S. and Windle, J. (2013). Bayesian inference for logistic models using pólya–gamma latent variables. Journal of the American Statistical Association , 108(504):1339--1349

    2013

  15. [23]

    and Zisserman, A

    Nilsback, M.-E. and Zisserman, A. (2008). Automated flower classification over a large number of classes. In Indian Conference on Computer Vision, Graphics & Image Processing

    2008

  16. [24]

    Oh, C., Kim, M., Lim, H., Park, J., Jeong, E., Cheng, Z.-Q., and Song, K. (2023). Towards calibrated robust fine-tuning of vision-language models. arXiv preprint arXiv:2311.01723

    2023 arXiv

  17. [25]

    M., Vedaldi, A., Zisserman, A., and Jawahar, C

    Parkhi, O. M., Vedaldi, A., Zisserman, A., and Jawahar, C. (2012). Cats and dogs. In IEEE Conference on Computer Vision and Pattern Recognition

    2012

  18. [26]

    W., Hallacy, C., Ramesh, A., Goh, G., Agarwal, S., Sastry, G., Askell, A., Mishkin, P., Clark, J., et al

    Radford, A., Kim, J. W., Hallacy, C., Ramesh, A., Goh, G., Agarwal, S., Sastry, G., Askell, A., Mishkin, P., Clark, J., et al. (2021). Learning transferable visual models from natural language supervision. In International conference on machine learning , pages 8748--8763. PMLR

    2021

  19. [27]

    and Zemel, R

    Snell, J. and Zemel, R. (2021). Bayesian few-shot classification with one-vs-each p \'o lya-gamma augmented gaussian processes. In International Conference on Learning Representations

    2021

  20. [28]

    R., and Shah, M

    Soomro, K., Zamir, A. R., and Shah, M. (2012a). A dataset of 101 human action classes from videos in the wild. Center for Research in Computer Vision

    2012

  21. [29]

    R., and Shah, M

    Soomro, K., Zamir, A. R., and Shah, M. (2012b). Ucf101: A dataset of 101 human actions classes from videos in the wild. ArXiv , abs/1212.0402

    2012 arXiv

  22. [30]

    Sun, Q., Fang, Y., Wu, L., Wang, X., and Cao, Y. (2023). Eva-clip: Improved training techniques for clip at scale. arXiv preprint arXiv:2303.15389

    2023 arXiv

  23. [31]

    Titsias RC AUEB, M. (2016). One-vs-each approximation to softmax for scalable estimation of probabilities. In Lee, D., Sugiyama, M., Luxburg, U., Guyon, I., and Garnett, R., editors, Advances in Neural Information Processing Systems , volume 29. Curran Associates, Inc

    2016

  24. [32]

    Veličković, P., Perivolaropoulos, C., Barbero, F., and Pascanu, R. (2024). softmax is not enough (for sharp out-of-distribution)

    2024

  25. [33]

    H., and Yang, L

    Wu, C.-E., Tian, Y., Yu, H., Wang, H., Morgado, P., Hu, Y. H., and Yang, L. (2023). Why is prompt tuning for vision-language models robust to noisy labels? In Proceedings of the IEEE/CVF International Conference on Computer Vision , pages 15488--15497

    2023

  26. [34]

    A., Oliva, A., and Torralba, A

    Xiao, J., Hays, J., Ehinger, K. A., Oliva, A., and Torralba, A. (2010a). Sun database: Large-scale scene recognition from abbey to zoo. 2010 IEEE Computer Society Conference on Computer Vision and Pattern Recognition , pages 3485--3492

    2010

  27. [35]

    A., Oliva, A., and Torralba, A

    Xiao, J., Hays, J., Ehinger, K. A., Oliva, A., and Torralba, A. (2010b). Sun database: Large-scale scene recognition from abbey to zoo. In IEEE Conference on Computer Vision and Pattern Recognition

    2010

  28. [36]

    Yang, Y., Ko, J., and Yun, S.-Y. (2024). Towards difficulty-agnostic efficient transfer learning for vision-language models. In Al-Onaizan, Y., Bansal, M., and Chen, Y.-N., editors, Proceedings of the 2024 Conference on Empirical Methods in Natural Language Processing , pages ...

    2024

  29. [37]

    Yao, H., Zhang, R., and Xu, C. (2023). Visual-language prompt tuning with knowledge-guided context optimization. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , pages 6757--6767

    2023

  30. [38]

    Yu, J., Wang, Z., Vasudevan, V., Yeung, L., Seyedhosseini, M., and Wu, Y. (2022). Coca: Contrastive captioners are image-text foundation models. arXiv preprint arXiv:2205.01917

    2022 arXiv

  31. [39]

    Zhai, X., Mustafa, B., Kolesnikov, A., and Beyer, L. (2023). Sigmoid loss for language image pre-training. arXiv preprint arXiv:2303.15343

    2023 arXiv

  32. [40]

    C., and Liu, Z

    Zhou, K., Yang, J., Loy, C. C., and Liu, Z. (2022a). Conditional prompt learning for vision-language models. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , pages 16816--16825

    2022

  33. [41]

    C., and Liu, Z

    Zhou, K., Yang, J., Loy, C. C., and Liu, Z. (2022b). Learning to prompt for vision-language models. International Journal of Computer Vision , 130(9):2337--2348

    2022

  34. [42]

    Zhu, B., Niu, Y., Han, Y., Wu, Y., and Zhang, H. (2022). Prompt-aligned gradient for prompt tuning. arXiv preprint arXiv:2205.14865

    2022 arXiv

Pith tools

Reviewed August 16, 2026 · model on record in the stance chip above.