Pith. sign in

REVIEW 4 major objections 5 minor 125 references

Test-Time Adaptation with Online Personalized Energy-Based Cache for Fine-Grained Video Expression Recognition

T0 review · 4 major / 5 minor · reviewed 2026-08-15 · deepseek-v4-flash

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

Pith's one-line read Frozen CLIP model personalizes video expression recognition per subject

desk verdict Strong empirical results on video FER test-time adaptation, but the energy-based sampler is effectively text-guided interpolation and the paper's central mechanism claim needs a control experiment. read the letter →

arxiv 2608.06467 v1 pith:GZTIOYEI submitted 2026-08-06 cs.CV

Masoumeh Sharafi , Muhammad Osama Zeeshan , Soufiane Belharbi , Alessandro Lameiras Koerich , Marco Pedersoli , Eric Granger This is my paper · ORCID
classification cs.CV
keywords test-timeadaptationcachepersonalizationenergy-basedmodelfacialexpressionrecognitionvideounderstandingCLIPstochasticgradientLangevindynamicssubject-specificdomainshift
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

This paper introduces Energy-Based Cache Personalization (EB-CaP), a test-time adaptation method for fine-grained facial expression recognition in videos. EB-CaP generates, for each unlabeled target video, class-specific prototype embeddings by running stochastic gradient Langevin dynamics chains that start from the current video's CLIP embedding and are guided by the energy $E_{\theta}(z,c)=-z^{\top}e_c$, where $e_c$ is the CLIP text embedding of expression class $c$. These sampled prototypes populate a personalized cache that is rebuilt for every video, alongside positive and negative caches of confident and uncertain target embeddings that are maintained per subject. A fused score combines the base CLIP score with similarities from the three caches. On BioVid, StressID, and BAH the method reports the best weighted average recall and F1 among compared optimization- and cache-based test-time adaptation methods, while keeping the recognition model frozen.

What carries the argument

The load-bearing object is the personalized sampled cache, built by a class-conditional energy-based sampler. The energy is the negative cosine-compatibility between an embedding and the class text embedding, $E_{\theta}(z,c)=-z^{\top}e_c$, and sampling is done by stochastic gradient Langevin dynamics, $z_{k+1}=z_k-\frac{\alpha}{2}\nabla_z E_\theta(z_k,c)+\sqrt{\alpha}\sigma\epsilon_k$, with each chain initialized from the current target video embedding and stopped early when the classifier's argmax is the conditioning class. This produces $m_s=3$ embeddings per class per video, replaces the usual need to accumulate target samples into class prototypes, and is re-run for every temporal window.

What would settle it

Use a labeled held-out set for the same target subjects. For each video, compute EB-CaP's sampled prototypes and check whether their nearest labeled frames belong to the conditioning class as often as the original video embedding's nearest frames do. If the sampled prototypes match the conditioning class less often, the sampler is drifting toward text directions rather than subject-specific evidence.

Watch

Extended reading notes

Core claim

EB-CaP claims that the cold-start problem of cache-based test-time adaptation can be solved by synthesizing class prototypes on demand rather than collecting them from observed target data. Starting each SGLD chain from the current unlabeled video representation and stopping as soon as the classifier's argmax becomes the conditioning class produces personalized, class-conditional embeddings for every class from the first video on. The paper further claims that combining these synthesized prototypes with a positive cache of reliable target embeddings and a negative cache of uncertain ones, both filtered by an adaptive entropy gate and a diversity gate, yields subject-specific adaptation that outperforms prompt-optimization methods like TPT and PromptAlign and cache methods like TDA, DPE, ReTA, and T3AL on the three benchmarks. The central contribution is that no source samples, source prototypes, target labels, or parameter updates are needed at test time.

Load-bearing premise

The method assumes that pushing the target video embedding toward the CLIP class-text embedding, with early stopping at the argmax switch, produces sampled embeddings that remain representative of the target subject's actual expression rather than drifting toward a generic class direction.

Editorial extensions

If this is right

  • If the central claim holds, test-time adaptation for video facial expression recognition can be done without any backpropagation through the recognition model, removing the instability and latency of entropy-minimization prompt tuning.
  • A personalized cache covering all classes is available from the very first video of a subject, so rarely observed classes do not have to wait for target samples to accumulate.
  • Because only pretrained CLIP and a fixed temporal encoder are used, the method can be dropped into existing CLIP-based FER pipelines without retraining or storing source prototypes.
  • The reported per-subject gains on BioVid and StressID suggest that subject-level evaluation is the right lens for comparing personalization methods, since averages can hide large differences at the individual level.

Reading between the lines

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

  • The same energy-guided sampling idea could be applied to other fine-grained video tasks, such as action-unit detection or pain intensity estimation, where class text embeddings exist and subject-specific shifts dominate; the paper does not test this.
  • One unresolved question the paper leaves open is whether the sampled prototypes are truly subject-specific or merely class-text directions with noise; a direct comparison of sampled embeddings against source-prototype embeddings from the same classes would clarify the mechanism.
  • The early-stopping criterion (stop when the argmax becomes c) is cheap but asymmetric: classes that are already predicted for the starting video will stop almost immediately, while others run longer, and this asymmetry may bias cache diversity; a testable variant would impose a minimum chain length.
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 EB-CaP, a test-time adaptation method for fine-grained video facial expression recognition. For each incoming unlabeled video, EB-CaP samples class-conditioned embeddings from the current video representation using an energy-based model whose energy is the negative cosine similarity to CLIP class text embeddings, and stores these in a personalized sampled cache. In parallel, positive and negative target caches accumulate reliable and uncertain embeddings under an adaptive entropy gate and a diversity gate. Final predictions fuse CLIP scores with similarities to the sampled, positive, and negative caches. Experiments on BioVid, StressID, and BAH report improved WAR and F1 over several optimization- and cache-based TTA baselines, with lower memory usage, and the authors provide subject-level results and an ablation study.

Significance. If the reported results and the proposed mechanism held, EB-CaP would be a practically attractive TTA method for video FER: it avoids test-time parameter updates, does not store source prototypes, and the released code and subject-level tables are useful contributions. The efficiency measurements in Table 5 and the per-subject results in Tables 2-4 are concrete and go beyond average-only reporting. However, the central mechanism is not currently established: the energy-based sampler described by Eqs. (1)-(2) is mathematically equivalent to a noisy spherical interpolation toward CLIP text embeddings, so the claim that it produces personalized class-conditional prototypes is unsupported. In addition, the main ablation compares against a zero-shot CLIP baseline rather than the actual frozen source-trained model, and the statistical evidence covers only BioVid WAR. These issues are load-bearing for the paper's central claims.

major comments (4)
  1. [Personalized Sampled Cache, Eqs. (1)-(2)] The proposed energy-based sampler does not depend on the source-trained recognition model and reduces to text-guided interpolation. In Eq. (1), E_theta(z,c) = -z^T e_c, where e_c is the fixed CLIP text embedding, so the SGLD update in Eq. (2) is z_{k+1} = z_k + (alpha/2) e_c + sqrt(alpha sigma) epsilon_k. After l2 normalization each chain is a noisy spherical interpolation from z_tgt toward e_c, and the early-stopping rule `argmax_{c'} z_k^T e_{c'} = c` simply places the sample on a CLIP decision boundary. The only subject-specific information is the initialization z_tgt; no source-trained class-conditional distribution is used. Since the base CLIP score s_tgt also uses the same e_c, the sampled cache is not an independent source of evidence. The paper should compare against a deterministic closed-form interpolation z = normalize(z_tgt + beta e_c) with the same stopping rule, and should either use a source-dependent energy function or explicitly reframe the method as text-guided interpolation. As written, the claimed personalization mechanism is not demonstrated.
  2. [Figure 3 and Table 1] The ablation baseline is inconsistent with the actual base model. Figure 3 reports a 'CLIP baseline' WAR of 50.0, which is the zero-shot CLIP value in Table 1, whereas EB-CaP uses a source-trained temporal encoder whose frozen model achieves 69.7 WAR in Table 1. The correct control for the cache ablations is the frozen source-trained model (69.7), not zero-shot CLIP. Without this control, the contribution of the sampled cache (77.9) and target cache (78.6) is overstated, and the apparent gain from 50.0 to 81.0 conflates source training with test-time adaptation. Please rerun the ablation against the correct frozen baseline and report the resulting increments.
  3. [Results, 'Comparison with State-of-the-Art Methods' and Tables 1-4] The claim that EB-CaP outperforms state-of-the-art TTA methods on all three datasets is not fully supported by the reported statistics. On BAH the improvement over T3AL is 1.0 WAR (68.9 vs 67.9) and 0.3 F1 (41.0 vs 40.7), and on several individual subjects the method is not best. No error bars or confidence intervals are provided, and the only significance test is a Wilcoxon signed-rank test on BioVid WAR. To support the cross-dataset claim, the authors should report significance tests (or error bars) for all three datasets and for both metrics, and should discuss the BAH case explicitly in terms of effect size.
  4. [Implementation Details and Ablation Studies] The main SGLD hyperparameters (step size alpha, noise scale sigma, and maximum number of iterations) are set by hand, and the provided sensitivity analysis covers only cache capacities (Figure 5) and fusion weights (supplementary Table 6). Because the sampler is the central contribution, the paper should include a sensitivity study for alpha, sigma, and the iteration limit to show that the reported gains are not artifacts of a particular SGLD configuration. This is especially important given that the sampler reduces to interpolation, where alpha and the stopping rule determine the distance traveled toward e_c.
minor comments (5)
  1. [Table 1] The labeling of the rows is confusing: the second row appears to denote a fine-tuned CLIP model, but the footnote 'CLIP-ViT-B/32† denotes full CLIP fine-tuning' does not visually match any row in the table. Please clarify which rows are zero-shot, frozen source-trained, and fully fine-tuned.
  2. [Eq. (3) and supplementary Table 6] Equation (3) writes the fused score as s_fuse = s_tgt + s_s + s_p - s_n, but the supplementary sensitivity analysis uses weights lambda_s, lambda_p, lambda_n. The main text should present the weighted equation or explicitly state that all weights are set to 1.
  3. [Abstract and Eq. (1)] The abstract says the energy relies only on pretrained CLIP, while Eq. (1) attributes the energy to a 'frozen recognition model parameterized by theta.' Since the energy has no dependence on theta, the notation and wording should be corrected to avoid implying source-model involvement.
  4. [Section 'Target Caches'] The notation m_p and m_n is used inconsistently: it is first described as the current number of entries per class and later as a fixed capacity. Please distinguish the capacity constant from the current count.
  5. [References] The reference for Sharafi et al. 2026a has a garbled author list ('Belharbi, M. O., S. Zeeshan'); please correct it.

Circularity Check

1 steps flagged · score 6.0 of 10

The EB-CaP sampled-cache score is constructed from the same CLIP text-similarity quantity as the base prediction, so the claimed personalized prototypes supply limited independent evidence.

  1. self definitional [Section 'TTA through Cache Personalization', Eqs. (1)-(3) and 'Cache Retrieval and Final Prediction']
    "Eθ(z,c) =−z ⊤ec, (1) ... Once the sampled embedding is predicted as the conditioning class c, the chain is stopped ... sampling stops when arg max c′∈C ez⊤ k ec′ = c ... sfuse =s tgt +ss +sp−sn, (3)"

    The base CLIP score is s_tgt(c)=η cos(z_tgt,e_c). The sampled-cache score s_s(c) aggregates cosine similarities between z_tgt and samples e_zc,j produced by Eq. (2). Because the SGLD gradient is +e_c and the chain is l2-normalized each step, every sample follows a noisy spherical path from z_tgt toward e_c, and the stopping rule forces the endpoint into the CLIP class-c cone. Therefore z_tgt^T e_zc,j is approximately a scalar multiple of z_tgt^T e_c, i.e., of the base CLIP score, plus noise. The sampled-cache score thus re-weights the same text-similarity evidence already used for the base prediction rather than providing independent subject-specific evidence; the only subject-specific content, the initialization z_tgt, already enters s_tgt.

full rationale

The derivation chain for the sampled cache is self-contained, but the claimed personalization mechanism reduces to CLIP text similarity. Eq. (1)-(2) define an energy that depends only on the CLIP text embedding e_c, and the sampled embeddings are initialized from the target representation z_tgt and pushed toward e_c. Because the base CLIP score s_tgt is also cosine similarity to e_c, the sampled-cache scores are strongly correlated with s_tgt. The paper does not demonstrate that the SGLD trajectory provides information beyond interpolation toward the class text embedding. Therefore the central claim that EB-CaP generates subject-specific class-conditional prototypes is partially circular: the sampled-cache evidence is derived from the same class-text similarity that defines the base prediction. However, the positive and negative caches are built from target data, so the method is not fully circular; score 6 reflects this partial reduction.

Assumptions & free parameters 8 free parameters · 4 assumptions · 0 invented entities

The central method depends on several hand-set hyperparameters and domain assumptions about the CLIP embedding space. The most consequential is that energy-based movement toward CLIP text embeddings yields subject-specific prototypes; this is plausible but untested in the paper. No fundamentally new physical or mathematical entities are introduced.

free parameters (8)
  • SGLD step size alpha = 0.01
    Set by hand in Implementation Details; no sensitivity analysis reported.
  • SGLD noise scale sigma = 0.1
    Set by hand; no sensitivity analysis reported.
  • Max SGLD iterations = 20
    Set by hand; chains may stop earlier once the conditioning class is predicted.
  • Sampled cache size ms = 3 per class
    Selected via ablation on BioVid (Figure 5).
  • Positive cache capacity mp = 5 per class
    Selected via ablation on BioVid (Figure 5).
  • Negative cache capacity mn = 4 per class
    Selected via ablation on BioVid (Figure 5).
  • Entropy thresholds for warm-up = 0.5 positive, 0.8 negative
    Fixed for the first five temporal representations of each video; chosen by hand.
  • Fusion weights lambda_s, lambda_p, lambda_n = 1, 1, 1
    Validated via sensitivity analysis on BioVid (Table 6); varying 0.5 to 1.5 changes WAR by less than 1 point.
assumptions (4)
  • domain assumption CLIP text embeddings e_c are valid class centers for fine-grained expression classes in the CLIP embedding space.
    Eq. (1) defines the energy as -z^T e_c, so the sampler assumes moving toward e_c yields class-conditional prototypes. The paper does not validate the class-text directions for subject-specific expression appearance.
  • domain assumption SGLD with early stopping produces sampled embeddings that remain representative of the target subject's expression while being class-conditional.
    Eq. (2) and the stopping rule assume the refined embedding is a valid prototype for the conditioning class; no analysis of distribution shift from the target video representation is provided.
  • domain assumption Predictive entropy from the refined probabilities is a reliable indicator of pseudo-label correctness for cache insertion.
    The adaptive entropy gate uses entropy thresholds to accept positive and negative cache entries; the paper provides no evidence that low entropy correlates with correct pseudo-labels under subject shift.
  • standard math Cosine similarity in the frozen CLIP embedding space is an appropriate kernel for comparing video representations to cached prototypes.
    Used throughout retrieval and fusion; a standard assumption in CLIP-based methods.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Test-Time Adaptation with Online Personalized Energy-Based Cache for Fine-Grained Video Expression Recognition." pith.science (2026). https://pith.science/paper/GZTIOYEI

@misc{pith2026260806467,
  author       = {Pith},
  title        = {Pith review of: Test-Time Adaptation with Online Personalized Energy-Based Cache for Fine-Grained Video Expression Recognition},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/GZTIOYEI}},
  note         = {Machine review of arXiv:2608.06467}
}
read the original abstract

Facial expression recognition (FER) in videos is challenging because models must identify subtle, temporally evolving affective states that vary across individuals. Although vision-language models provide transferable visual-semantic representations, models trained on subject-independent data often degrade under subject-specific distribution shifts at inference time. Existing test-time adaptation (TTA) methods commonly update model parameters during inference, increasing computational cost and latency. Cache-based methods avoid parameter updates, but they usually require enough target samples to form reliable class prototypes, which is difficult early in adaptation and for rarely observed classes. We introduce Energy-Based Cache Personalization (EB-CaP), a subject-based online TTA method for video FER that generates class-specific prototypes personalized to each target video. EB-CaP uses a lightweight energy-based model to sample prototypes from the current unlabeled video and populate a personalized cache online, without accumulating large amounts of target data or storing diverse source prototypes. Its energy function relies only on pretrained CLIP: similarities between the target video embedding and class text embeddings guide prototype sampling. In parallel, positive and negative caches store reliable and uncertain target embeddings. An adaptive entropy gate controls cache updates according to the evolving confidence distribution, while a diversity gate limits redundant samples. Final predictions combine cache-derived scores with the current CLIP scores. Experiments on BioVid, StressID, and BAH show that EB-CaP outperforms state-of-the-art TTA methods while maintaining low computational and memory overhead. Code is available at https://github.com/MasoumehSharafi/EB-CaP.

Figures

Figures reproduced from arXiv: 2608.06467 by the authors.

Figure 1
Figure 1. Cache-based TTA methods from target subject data. [PITH_FULL_IMAGE:figures/full_fig_p001_1.png] view at source ↗
Figure 2
Figure 2. Overview of EB-CaP for online TTA. Frozen CLIP encoders and a temporal encoder produce the current video representation and CLIP class scores. An energy-based model (EBM) samples class-conditioned embeddings initialized from the target video to construct a personalized sampled cache covering all classes. In parallel, adaptive entropy and diversity gates control updates to the positive and negative target caches. Fin… view at source ↗
Figure 3
Figure 3. Cache design on BioVid data. Results compare the CLIP baseline, personalized sampled cache, target cache (positive and negative), and complete EB-CaP method in terms of WAR. Entropy Diversity Both 0 20 40 60 80 10.4 22.4 5.9 Gate configuration Pass rate (%) Pass rate 74 76 78 80 82 79.1 78 81 WAR (%) WAR [PITH_FULL_IMAGE:figures/full_fig_p006_3.png] view at source ↗
Figures from the paper (2 more)
Figure 4
Figure 4. Figure 4: Adaptive entropy and diversity gates on BioVid. Bars report the percentage of candidates admitted by each gate configuration, while points show the corresponding WAR. “Both” indicates that both gates are jointly applied. research (Zhao et al. 2025), the prompt template…
Figure 5
Figure 5. Figure 5: Sensitivity analysis of the sampled, positive, and [PITH_FULL_IMAGE:figures/full_fig_p007_5.png]

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

125 extracted references · 40 canonical work pages

  1. [1]

    H.; Hussein, N.; Khattak, M

    Abdul Samadh, J.; Gani, M. H.; Hussein, N.; Khattak, M. U.; Naseer, M. M.; Shahbaz Khan, F.; and Khan, S. H. 2023. Align your prompts: Test-time prompting with distribution alignment for zero-shot generalization. Advances in Neural Information Processing Systems, 36: 80396--80413

    2023

  2. [2]

    A.; and D'Mello, S

    Calvo, R. A.; and D'Mello, S. 2010. Affect detection: An interdisciplinary review of models, methods, and their applications. IEEE Transactions on Affective Computing, 1(1): 18--37

    2010

  3. [3]

    A.; et al

    Chaptoukaev, H.; Strizhkova, V.; Panariello, M.; Dalpaos, B.; Reka, A.; Manera, V.; Th \"u mmler, S.; Ismailova, E.; Todisco, M.; Zuluaga, M. A.; et al. 2023. Stressid: a multimodal dataset for stress identification. Advances in Neural Information Processing Systems, 36: 29798--29811

    2023

  4. [4]

    Chen, S.; Chen, Y.; Yin, X.; Liu, X.; Lai, H.; and Zhang, T. 2025 a . PAF: Prototype Adaptive Fusion for Test-Time Adaptation of Vision-Language Models. In Proceedings of the 33rd ACM International Conference on Multimedia, 3007--3016

    2025

  5. [5]

    Chen, X.; Zhai, H.; Zhang, C.; Shi, X.; and Li, R. 2025 b . Multi-cache enhanced prototype learning for test-time generalization of vision-language models. In Proceedings of the IEEE/CVF International Conference on Computer Vision, 2281--2291

    2025

  6. [6]

    C.; Granger, E.; and Hadid, A

    De Melo, W. C.; Granger, E.; and Hadid, A. 2020. A deep multiscale spatiotemporal network for assessing depression from facial dynamics. IEEE transactions on affective computing, 13(3): 1581--1592

    2020

  7. [7]

    Duvenaud, D.; Wang, J.; Jacobsen, J.; Swersky, K.; Norouzi, M.; and Grathwohl, W. 2020. Your classifier is secretly an energy based model and you should treat it like one. ICLR 2020

    2020

  8. [8]

    Feng, C.-M.; Yu, K.; Liu, Y.; Khan, S.; and Zuo, W. 2023. Diverse data augmentation with diffusions for effective test-time prompt tuning. In Proceedings of the IEEE/CVF International Conference on Computer Vision, 2704--2714

    2023

Show all 125 references
  1. [9]

    M.; and Patras, I

    Foteinopoulou, N. M.; and Patras, I. 2024. Emoclip: A vision-language method for zero-shot video facial expression recognition. In 2024 IEEE 18th International Conference on Automatic Face and Gesture Recognition (FG), 1--10. IEEE

    2024

  2. [10]

    Gao, P.; Geng, S.; Zhang, R.; Ma, T.; Fang, R.; Zhang, Y.; Li, H.; and Qiao, Y. 2024. Clip-adapter: Better vision-language models with feature adapters. International Journal of Computer Vision, 132(2): 581--595

    2024

  3. [13]

    Guan, Z.; Yucan, Z.; Liu, W.; and Gu, X. 2025. Statistics Caching Test-Time Adaptation for Vision-Language Models. In The Thirty-ninth Annual Conference on Neural Information Processing Systems

    2025

  4. [14]

    N.; and Wang, Z

    Huang, F.; Jiang, J.; Jiang, Q.; Li, H.; Khan, F. N.; and Wang, Z. 2025. COSMIC: Clique-Oriented Semantic Multi-space Integration for Robust CLIP Test-Time Adaptation. In Proceedings of the Computer Vision and Pattern Recognition Conference, 9772--9781

    2025

  5. [15]

    Karmanov, A.; Guan, D.; Lu, S.; El Saddik, A.; and Xing, E. 2024. Efficient test-time adaptation of vision-language models. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, 14162--14171

    2024

  6. [17]

    Li, H.; Niu, H.; Zhu, Z.; and Zhao, F. 2024. Cliper: A unified vision-language framework for in-the-wild facial expression recognition. In 2024 IEEE International Conference on Multimedia and Expo (ICME), 1--6. IEEE

    2024

  7. [18]

    Li, J.; Li, D.; Savarese, S.; and Hoi, S. 2023. Blip-2: Bootstrapping language-image pre-training with frozen image encoders and large language models. In International conference on machine learning, 19730--19742. PMLR

    2023

  8. [20]

    Liang, Y.; Chen, H.; Xiong, Y.; Zhou, Z.; Lyu, M.; Lin, Z.; Niu, S.; Zhao, S.; Han, J.; and Ding, G. 2025. Advancing Reliable Test-Time Adaptation of Vision-Language Models under Visual Variations. In Proceedings of the 33rd ACM International Conference on Multimedia, 4788--4797

    2025

  9. [21]

    Liberatori, B.; Conti, A.; Rota, P.; Wang, Y.; and Ricci, E. 2024. Test-time zero-shot temporal action localization. In Proceedings of the IEEE/CVF conference on computer vision and pattern recognition, 18720--18729

    2024

  10. [22]

    Ma, F.; He, Y.; Sun, B.; and Li, S. 2025. Multimodal Prompt Alignment for Facial Expression Recognition. In IEEE/CVF International Conference on Computer Vision (ICCV)

    2025

  11. [24]

    Ni, B.; Peng, H.; Chen, M.; Zhang, S.; Meng, G.; Fu, J.; Xiang, S.; and Ling, H. 2022. Expanding Language-Image Pretrained Models for General Video Recognition. In European Conference on Computer Vision, 1--18

    2022

  12. [25]

    Osowiechi, D.; Noori, M.; Vargas Hakim, G.; Yazdanpanah, M.; Bahri, A.; Cheraghalikhani, M.; Dastani, S.; Beizaee, F.; Ayed, I.; and Desrosiers, C. 2024. WATT: Weight average test time adaptation of CLIP. Advances in neural information processing systems, 37: 48015--48044

    2024

  13. [26]

    Pu, J.; and Nie, X. 2023. Convolutional channel attentional facial expression recognition network and its application in human--computer interaction. IEEE Access, 11: 129412--129424

    2023

  14. [27]

    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, 8748--8763. PmLR

    2021

  15. [28]

    Sharafi, M.; Belharbi, M. O., S. Zeeshan; Ben Salem, H.; Etemad, A.; Koerich, A.; Pedersoli, M.; Bacon, S.; and Granger, E. 2026 a . Personalized Feature Translation for Expression Recognition: An Efficient Source-Free Domain Adaptation Method. In ICLR

    2026

  16. [29]

    O.; Belharbi, S.; Koerich, A

    Sharafi, M.; Ollivier, E.; Zeeshan, M. O.; Belharbi, S.; Koerich, A. L.; Pedersoli, M.; Bacon, S.; and Granger, E. 2025. Disentangled source-free personalization for facial expression recognition with neutral target data. In 2025 IEEE 19th International Conference on Automatic...

    2025

  17. [31]

    Shu, M.; Nie, W.; Huang, D.-A.; and Yu, Z. 2022. Test-time prompt tuning for zero-shot generalization in vision-language models. Advances in Neural Information Processing Systems, 35: 14274--14289

    2022

  18. [32]

    Shu, Y.; Guo, X.; Wu, J.; Wang, X.; Wang, J.; and Long, M. 2023. Clipood: Generalizing clip to out-of-distributions. In International conference on machine learning, 31716--31731. PMLR

    2023

  19. [33]

    Sun, Y.; Wang, X.; Liu, Z.; Miller, J.; Efros, A.; and Hardt, M. 2020. Test-time training with self-supervision for generalization under distribution shifts. In International conference on machine learning, 9229--9248. PMLR

    2020

  20. [34]

    Tu, W.; Deng, W.; and Gedeon, T. 2023. A closer look at the robustness of contrastive language-image pre-training (clip). Advances in Neural Information Processing Systems, 36: 13678--13691

    2023

  21. [35]

    C.; Werner, P.; Al-Hamadi, A.; Crawcour, S.; Andrade, A

    Walter, S.; Gruss, S.; Ehleiter, H.; Tan, J.; Traue, H. C.; Werner, P.; Al-Hamadi, A.; Crawcour, S.; Andrade, A. O.; and da Silva, G. M. 2013. The biovid heat pain database data for the advancement and systematic validation of an automated pain recognition system. In 2013 IEEE...

    2013

  22. [36]

    Wang, Q.; Fink, O.; Van Gool, L.; and Dai, D. 2022. Continual test-time domain adaptation. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, 7201--7211

    2022

  23. [37]

    Welling, M.; and Teh, Y. W. 2011. Bayesian learning via stochastic gradient Langevin dynamics. In Proceedings of the 28th international conference on machine learning (ICML-11), 681--688

    2011

  24. [38]

    Yi, C.; Ren, L.; Zhan, D.-C.; and Ye, H.-J. 2024. Leveraging cross-modal neighbor representation for improved clip classification. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, 27402--27411

    2024

  25. [41]

    Yuan, L.; Xie, B.; and Li, S. 2023. Robust test-time adaptation in dynamic scenarios. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, 15922--15932

    2023

  26. [42]

    O.; Aslam, M

    Zeeshan, M. O.; Aslam, M. H.; Belharbi, S.; Koerich, A. L.; Pedersoli, M.; Bacon, S.; and Granger, E. 2024. Subject-based domain adaptation for facial expression recognition. In 2024 IEEE 18th International Conference on Automatic Face and Gesture Recognition (FG), 1--10. IEEE

    2024

  27. [44]

    O.; Pedersoli, M.; Koerich, A

    Zeeshan, M. O.; Pedersoli, M.; Koerich, A. L.; and Granger, E. 2025 b . Progressive multi-source domain adaptation for personalized facial expression recognition. IEEE Transactions on Affective Computing

    2025

  28. [46]

    Zhai, H.; Chen, X.; Zhang, C.; Sha, T.; and Li, R. 2025. Mitigating cache noise in test-time adaptation for large vision-language models. In 2025 IEEE International Conference on Multimedia and Expo (ICME), 1--6. IEEE

    2025

  29. [47]

    Zhang, C.; Stepputtis, S.; Sycara, K.; and Xie, Y. 2024. Dual prototype evolving for test-time generalization of vision-language models. Advances in Neural Information Processing Systems, 37: 32111--32136

    2024

  30. [48]

    Zhang, D.-C.; Zhou, Z.; and Li, Y.-F. 2024. Robust test-time adaptation for zero-shot prompt tuning. In Proceedings of the AAAI Conference on Artificial Intelligence, volume 38, 16714--16722

    2024

  31. [49]

    Zhang, R.; Zhang, W.; Fang, R.; Gao, P.; Li, K.; Dai, J.; Qiao, Y.; and Li, H. 2022. Tip-adapter: Training-free adaption of clip for few-shot classification. In European conference on computer vision, 493--510. Springer

    2022

  32. [50]

    Zhao, Z.; Cao, Y.; Gong, S.; and Patras, I. 2025. Enhancing zero-shot facial expression recognition by llm knowledge transfer. In 2025 IEEE/CVF Winter Conference on Applications of Computer Vision (WACV)

    2025

  33. [51]

    Zhao, Z.; and Patras, I. 2023. Prompting Visual-Language Models for Dynamic Facial Expression Recognition. In BMVC

    2023

  34. [52]

    Zhu, X.; Wang, S.; Zhu, B.; Li, M.; Li, Y.; Fang, J.; Wang, Z.; Wang, D.; and Zhang, H. 2025. Dynamic multimodal prototype learning in vision-language models. In Proceedings of the IEEE/CVF international conference on computer vision, 2501--2511

    2025

  35. [53]

    Communication, Simulation, and Intelligent Agents: Implications of Personal Intelligent Machines for Medical Education

    Clancey, William J. Communication, Simulation, and Intelligent Agents: Implications of Personal Intelligent Machines for Medical Education. Proceedings of the Eighth International Joint Conference on Artificial Intelligence (IJCAI-83)

  36. [54]

    Classification Problem Solving

    Clancey, William J. Classification Problem Solving. Proceedings of the Fourth National Conference on Artificial Intelligence

  37. [55]

    , title =

    Robinson, Arthur L. , title =. 1980 , doi =. https://science.sciencemag.org/content/208/4447/1019.full.pdf , journal =

    1980

  38. [56]

    New Ways to Make Microcircuits Smaller---Duplicate Entry

    Robinson, Arthur L. New Ways to Make Microcircuits Smaller---Duplicate Entry. Science

  39. [57]

    Clancey and Glenn Rennels , abstract =

    Diane Warner Hasling and William J. Clancey and Glenn Rennels , abstract =. Strategic explanations for a diagnostic consultation system , journal =. 1984 , issn =. doi:https://doi.org/10.1016/S0020-7373(84)80003-6 , url =

    1984 doi

  40. [58]

    and Rennels, Glenn R

    Hasling, Diane Warner and Clancey, William J. and Rennels, Glenn R. and Test, Thomas. Strategic Explanations in Consultation---Duplicate. The International Journal of Man-Machine Studies

  41. [59]

    Poligon: A System for Parallel Problem Solving

    Rice, James. Poligon: A System for Parallel Problem Solving

  42. [60]

    Transfer of Rule-Based Expertise through a Tutorial Dialogue

    Clancey, William J. Transfer of Rule-Based Expertise through a Tutorial Dialogue

  43. [61]

    The Engineering of Qualitative Models

    Clancey, William J. The Engineering of Qualitative Models

  44. [62]

    2023 , eprint=

    Attention Is All You Need , author=. 2023 , eprint=

    2023

  45. [63]

    Pluto: The 'Other' Red Planet

    NASA. Pluto: The 'Other' Red Planet

  46. [64]

    IEEE Access , volume=

    Convolutional channel attentional facial expression recognition network and its application in human--computer interaction , author=. IEEE Access , volume=. 2023 , publisher=

    2023

  47. [65]

    IEEE Transactions on Affective Computing , volume=

    Affect detection: An interdisciplinary review of models, methods, and their applications , author=. IEEE Transactions on Affective Computing , volume=

  48. [66]

    arXiv preprint arXiv:2006.10726 , year=

    Tent: Fully test-time adaptation by entropy minimization , author=. arXiv preprint arXiv:2006.10726 , year=

    2006 arXiv

  49. [67]

    International conference on machine learning , pages=

    Learning transferable visual models from natural language supervision , author=. International conference on machine learning , pages=. 2021 , organization=

    2021

  50. [68]

    2024 IEEE International Conference on Multimedia and Expo (ICME) , pages=

    Cliper: A unified vision-language framework for in-the-wild facial expression recognition , author=. 2024 IEEE International Conference on Multimedia and Expo (ICME) , pages=. 2024 , organization=

    2024

  51. [69]

    arXiv preprint arXiv:2205.01917 , year=

    Coca: Contrastive captioners are image-text foundation models , author=. arXiv preprint arXiv:2205.01917 , year=

    arXiv

  52. [70]

    Proceedings of the IEEE/CVF international conference on computer vision , pages=

    Dynamic multimodal prototype learning in vision-language models , author=. Proceedings of the IEEE/CVF international conference on computer vision , pages=

  53. [71]

    2025 IEEE International Conference on Multimedia and Expo (ICME) , pages=

    Mitigating cache noise in test-time adaptation for large vision-language models , author=. 2025 IEEE International Conference on Multimedia and Expo (ICME) , pages=. 2025 , organization=

    2025

  54. [72]

    Proceedings of the IEEE/CVF International Conference on Computer Vision , pages=

    Multi-cache enhanced prototype learning for test-time generalization of vision-language models , author=. Proceedings of the IEEE/CVF International Conference on Computer Vision , pages=

  55. [73]

    Proceedings of the 33rd ACM International Conference on Multimedia , pages=

    PAF: Prototype Adaptive Fusion for Test-Time Adaptation of Vision-Language Models , author=. Proceedings of the 33rd ACM International Conference on Multimedia , pages=

  56. [74]

    The Thirty-ninth Annual Conference on Neural Information Processing Systems , year=

    Statistics Caching Test-Time Adaptation for Vision-Language Models , author=. The Thirty-ninth Annual Conference on Neural Information Processing Systems , year=

  57. [75]

    arXiv preprint arXiv:2110.05208 , year=

    Supervision exists everywhere: A data efficient contrastive language-image pre-training paradigm , author=. arXiv preprint arXiv:2110.05208 , year=

    arXiv

  58. [76]

    International conference on machine learning , pages=

    Blip-2: Bootstrapping language-image pre-training with frozen image encoders and large language models , author=. International conference on machine learning , pages=. 2023 , organization=

    2023

  59. [77]

    Proceedings of the IEEE/CVF International Conference on Computer Vision , pages=

    Diverse data augmentation with diffusions for effective test-time prompt tuning , author=. Proceedings of the IEEE/CVF International Conference on Computer Vision , pages=

  60. [78]

    International conference on machine learning , pages=

    Clipood: Generalizing clip to out-of-distributions , author=. International conference on machine learning , pages=. 2023 , organization=

    2023

  61. [79]

    Advances in Neural Information Processing Systems , volume=

    A closer look at the robustness of contrastive language-image pre-training (clip) , author=. Advances in Neural Information Processing Systems , volume=

  62. [80]

    Advances in Neural Information Processing Systems , volume=

    Align your prompts: Test-time prompting with distribution alignment for zero-shot generalization , author=. Advances in Neural Information Processing Systems , volume=

  63. [81]

    arXiv preprint arXiv:2403.14119 , year=

    C-tpt: Calibrated test-time prompt tuning for vision-language models via text feature dispersion , author=. arXiv preprint arXiv:2403.14119 , year=

    arXiv

  64. [82]

    Proceedings of the AAAI Conference on Artificial Intelligence , volume=

    Robust test-time adaptation for zero-shot prompt tuning , author=. Proceedings of the AAAI Conference on Artificial Intelligence , volume=

  65. [83]

    Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , pages=

    Efficient test-time adaptation of vision-language models , author=. Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , pages=

  66. [84]

    Advances in Neural Information Processing Systems , volume=

    Dual prototype evolving for test-time generalization of vision-language models , author=. Advances in Neural Information Processing Systems , volume=

  67. [85]

    Proceedings of the IEEE/CVF conference on computer vision and pattern recognition , pages=

    Dual memory networks: A versatile adaptation approach for vision-language models , author=. Proceedings of the IEEE/CVF conference on computer vision and pattern recognition , pages=

  68. [86]

    Proceedings of the Computer Vision and Pattern Recognition Conference , pages=

    COSMIC: Clique-Oriented Semantic Multi-space Integration for Robust CLIP Test-Time Adaptation , author=. Proceedings of the Computer Vision and Pattern Recognition Conference , pages=

  69. [87]

    2024 IEEE 18th International Conference on Automatic Face and Gesture Recognition (FG) , pages=

    Emoclip: A vision-language method for zero-shot video facial expression recognition , author=. 2024 IEEE 18th International Conference on Automatic Face and Gesture Recognition (FG) , pages=. 2024 , organization=

    2024

  70. [88]

    BMVC , year=

    Prompting Visual-Language Models for Dynamic Facial Expression Recognition , author=. BMVC , year=

  71. [89]

    IEEE/CVF International Conference on Computer Vision (ICCV) , year =

    Multimodal Prompt Alignment for Facial Expression Recognition , author =. IEEE/CVF International Conference on Computer Vision (ICCV) , year =

  72. [90]

    International conference on machine learning , pages=

    Test-time training with self-supervision for generalization under distribution shifts , author=. International conference on machine learning , pages=. 2020 , organization=

    2020

  73. [91]

    Advances in neural information processing systems , volume=

    Memo: Test time robustness via adaptation and augmentation , author=. Advances in neural information processing systems , volume=

  74. [92]

    International conference on machine learning , pages=

    Efficient test-time model adaptation without forgetting , author=. International conference on machine learning , pages=. 2022 , organization=

    2022

  75. [93]

    International Conference on Learning Representations (ICLR) , year =

    Towards Stable Test-Time Adaptation in Dynamic Wild World , author =. International Conference on Learning Representations (ICLR) , year =

  76. [94]

    Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , pages=

    Continual test-time domain adaptation , author=. Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , pages=

  77. [95]

    Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , pages=

    Robust test-time adaptation in dynamic scenarios , author=. Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , pages=

  78. [96]

    International Journal of Computer Vision , volume=

    Clip-adapter: Better vision-language models with feature adapters , author=. International Journal of Computer Vision , volume=. 2024 , publisher=

    2024

  79. [97]

    European conference on computer vision , pages=

    Tip-adapter: Training-free adaption of clip for few-shot classification , author=. European conference on computer vision , pages=. 2022 , organization=

    2022

  80. [98]

    Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , pages=

    Leveraging cross-modal neighbor representation for improved clip classification , author=. Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , pages=

  81. [99]

    Proceedings of the 33rd ACM International Conference on Multimedia , pages=

    Advancing Reliable Test-Time Adaptation of Vision-Language Models under Visual Variations , author=. Proceedings of the 33rd ACM International Conference on Multimedia , pages=

  82. [100]

    arXiv preprint arXiv:2507.14312 , year=

    Cliptta: Robust contrastive vision-language test-time adaptation , author=. arXiv preprint arXiv:2507.14312 , year=

  83. [101]

    Proceedings of the IEEE/CVF conference on computer vision and pattern recognition , pages=

    Badclip: Dual-embedding guided backdoor attack on multimodal contrastive learning , author=. Proceedings of the IEEE/CVF conference on computer vision and pattern recognition , pages=

  84. [102]

    Advances in neural information processing systems , volume=

    WATT: Weight average test time adaptation of CLIP , author=. Advances in neural information processing systems , volume=

  85. [103]

    European Conference on Computer Vision , pages=

    Expanding Language-Image Pretrained Models for General Video Recognition , author=. European Conference on Computer Vision , pages=

  86. [104]

    International Conference on Medical Image Computing and Computer-Assisted Intervention , pages=

    Source-free domain adaptation for medical image segmentation via prototype-anchored feature alignment and contrastive learning , author=. International Conference on Medical Image Computing and Computer-Assisted Intervention , pages=. 2023 , organization=

    2023

  87. [105]

    IEEE Transactions on Circuits and Systems for Video Technology , volume=

    VDM-DA: Virtual domain modeling for source data-free domain adaptation , author=. IEEE Transactions on Circuits and Systems for Video Technology , volume=. 2021 , publisher=

    2021

  88. [106]

    arXiv preprint arXiv:2106.15326 , year=

    Source-free domain adaptation via avatar prototype generation and adaptation , author=. arXiv preprint arXiv:2106.15326 , year=

    arXiv

  89. [107]

    ACM Transactions on Multimedia Computing, Communications and Applications , year=

    PE-CLIP: A Parameter-Efficient Fine-Tuning of Vision Language Models for Dynamic Facial Expression Recognition , author=. ACM Transactions on Multimedia Computing, Communications and Applications , year=

  90. [108]

    kdd , volume=

    A density-based algorithm for discovering clusters in large spatial databases with noise , author=. kdd , volume=

  91. [109]

    Journal of the American Statistical association , volume=

    Objective criteria for the evaluation of clustering methods , author=. Journal of the American Statistical association , volume=. 1971 , publisher=

    1971

  92. [110]

    Journal of classification , volume=

    Comparing partitions , author=. Journal of classification , volume=. 1985 , publisher=

    1985

  93. [111]

    Journal of Machine Learning Research , volume=

    Information Theoretic Measures for Clusterings Comparison: Variants, Properties, Normalization and Correction for Chance , author=. Journal of Machine Learning Research , volume=

  94. [112]

    Biocomputing 2002 , pages=

    A stability based method for discovering structure in clustered data , author=. Biocomputing 2002 , pages=. 2001 , publisher=

    2002

  95. [113]

    2013 IEEE international conference on cybernetics (CYBCO) , pages=

    The biovid heat pain database data for the advancement and systematic validation of an automated pain recognition system , author=. 2013 IEEE international conference on cybernetics (CYBCO) , pages=. 2013 , organization=

    2013

  96. [114]

    Advances in Neural Information Processing Systems , volume=

    Stressid: a multimodal dataset for stress identification , author=. Advances in Neural Information Processing Systems , volume=

  97. [115]

    arXiv preprint arXiv:2505.19328 , year=

    BAH Dataset for Ambivalence/Hesitancy Recognition in Videos for Behavioural Change , author=. arXiv preprint arXiv:2505.19328 , year=

    arXiv

  98. [116]

    2024 IEEE 18th International Conference on Automatic Face and Gesture Recognition (FG) , pages=

    Subject-based domain adaptation for facial expression recognition , author=. 2024 IEEE 18th International Conference on Automatic Face and Gesture Recognition (FG) , pages=. 2024 , organization=

    2024

  99. [117]

    Advances in Neural Information Processing Systems , volume=

    Test-time prompt tuning for zero-shot generalization in vision-language models , author=. Advances in Neural Information Processing Systems , volume=

  100. [118]

    Proceedings of the IEEE/CVF conference on computer vision and pattern recognition , pages=

    Test-time zero-shot temporal action localization , author=. Proceedings of the IEEE/CVF conference on computer vision and pattern recognition , pages=

  101. [119]

    In 2025 IEEE/CVF Winter Conference on Applications of Computer Vision (WACV) , author=

    Enhancing zero-shot facial expression recognition by llm knowledge transfer. In 2025 IEEE/CVF Winter Conference on Applications of Computer Vision (WACV) , author=. 2025 , publisher=

    2025

  102. [120]

    arXiv preprint arXiv:2508.12522 , year=

    MuSACo: Multimodal Subject-Specific Selection and Adaptation for Expression Recognition with Co-Training , author=. arXiv preprint arXiv:2508.12522 , year=

  103. [121]

    2025 IEEE 19th International Conference on Automatic Face and Gesture Recognition (FG) , year=

    Disentangled source-free personalization for facial expression recognition with neutral target data , author=. 2025 IEEE 19th International Conference on Automatic Face and Gesture Recognition (FG) , year=

    2025

  104. [122]

    ICLR , year=

    Personalized Feature Translation for Expression Recognition: An Efficient Source-Free Domain Adaptation Method , author=. ICLR , year=

  105. [123]

    IEEE Transactions on Affective Computing , year=

    Progressive multi-source domain adaptation for personalized facial expression recognition , author=. IEEE Transactions on Affective Computing , year=

  106. [124]

    Current biology , volume=

    Cultural confusions show that facial expressions are not universal , author=. Current biology , volume=. 2009 , publisher=

    2009

  107. [125]

    Proceedings of the National Academy of Sciences , volume=

    Facial expressions of emotion are not culturally universal , author=. Proceedings of the National Academy of Sciences , volume=. 2012 , publisher=

    2012

  108. [126]

    IEEE transactions on information theory , volume=

    Least squares quantization in PCM , author=. IEEE transactions on information theory , volume=. 1982 , publisher=

    1982

  109. [127]

    Some methods of classification and analysis of multivariate observations , author=. Proc. of 5th Berkeley Symposium on Math. Stat. and Prob. , pages=

  110. [128]

    arXiv preprint arXiv:2502.02618 , year=

    Deep learning-based facial expression recognition for the elderly: A systematic review , author=. arXiv preprint arXiv:2502.02618 , year=

    arXiv

  111. [129]

    arXiv preprint arXiv:2508.07570 , year=

    Adaptive Cache Enhancement for Test-Time Adaptation of Vision-Language Models , author=. arXiv preprint arXiv:2508.07570 , year=

  112. [130]

    International conference on machine learning , pages=

    On calibration of modern neural networks , author=. International conference on machine learning , pages=. 2017 , organization=

    2017

  113. [131]

    ICLR 2020 , year=

    Your classifier is secretly an energy based model and you should treat it like one , author=. ICLR 2020 , year=

    2020

  114. [132]

    IEEE transactions on affective computing , volume=

    A deep multiscale spatiotemporal network for assessing depression from facial dynamics , author=. IEEE transactions on affective computing , volume=. 2020 , publisher=

    2020

  115. [133]

    arXiv preprint arXiv:2603.27999 , year=

    Clip-autt: Test-time personalization with action unit prompting for fine-grained video emotion recognition , author=. arXiv preprint arXiv:2603.27999 , year=

    arXiv

  116. [134]

    arXiv preprint arXiv:2603.21309 , year=

    Test-Time Adaptation via Cache Personalization for Facial Expression Recognition in Videos , author=. arXiv preprint arXiv:2603.21309 , year=

    arXiv

  117. [135]

    Proceedings of the 28th international conference on machine learning (ICML-11) , pages=

    Bayesian learning via stochastic gradient Langevin dynamics , author=. Proceedings of the 28th international conference on machine learning (ICML-11) , pages=

Pith tools

Reviewed August 15, 2026 · model on record in the stance chip above.