Pith. sign in

REVIEW 3 major objections 4 minor 49 references

When Semantics Saturate or Emerge: Adaptation-Conditional Semantic Utility in Source-Free Cross-Domain Few-Shot Learning

T0 review · 3 major / 4 minor · reviewed 2026-08-10 · deepseek-v4-flash

Pith's one-line read The paper argues that detailed class descriptions do not have a fixed value across frozen and adapted visual encoders: their advantage can largely evaporate after visual LoRA, or appear only after it.

desk verdict A careful paired empirical study showing frozen prompt ranking flips after LoRA adaptation; the main caveat is that the 'detailed' prompts carry extra target-domain side information, so the semantics-versus-information-access question stays open. read the letter →

arxiv 2608.06673 v1 pith:NPHXHJD2 submitted 2026-08-07 cs.CV

classification cs.CV
keywords source-freecross-domainfew-shotlearningvision-languagemodelsLow-RankAdaptationpromptevaluationsemanticsaturationemergenceCLIPadaptation-conditionalutility
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

Source-free cross-domain few-shot learning often picks language descriptions by testing them on a frozen model, on the assumption that a prompt that scores well before adaptation will also be a good anchor for adaptation. This paper tests that assumption with a strictly paired experiment: for the same episodes, only the text view changes, while visual Low-Rank Adaptation (LoRA), the support set, and the optimization stream stay fixed. It finds that the usefulness of detailed descriptions is adaptation-conditional. On EuroSAT and CropDisease, detailed descriptions give a large frozen-model advantage that mostly disappears after LoRA; on ISIC and ChestX, they give no frozen-model advantage but become the better anchor after LoRA. The paper concludes that frozen-model prompt quality is an incomplete proxy for adaptation-anchor quality, and that prompt evaluation should happen on both sides of the adaptation boundary.

What carries the argument

The apparatus is the paired text-view protocol plus a fixed visual LoRA probe. Each episode is run twice with identical classes, support/query images, initialization, data stream, and update schedule; only the text anchor set used by the support loss changes, between a class-name template and fixed class-level descriptions. The argument is carried by the two utility quantities $\Delta_0$ and $\Delta_L$, by checkpoint-dependent trajectories $\Delta_t$, and by the exact sample-level identity $\Delta_L = X_D - X_B$, where $X_D$ and $X_B$ are the fractions of query samples correct exclusively under Detailed-LoRA and Base-LoRA. Because the comparison is paired, any utility difference is attributable to the text view, and because LoRA is trained separately against each view, the frozen and adapted readings can diverge. The shuffled-semantic control preserves the text-embedding multiset while destroying class–description correspondence, isolating genuine semantics from mere text length or codebook geometry.

What would settle it

Collect a battery of, say, twenty text-view pairs across domains and backbones under the same paired protocol, and compute $\Delta_0$ and $\Delta_L$ for each. If the two quantities are perfectly rank-ordered and no pair falls in the emergence cell ($\Delta_0 \le 0 < \Delta_L$), or if the saturation regime is absent everywhere, the claim that zero-shot prompt quality is an unreliable proxy for adaptation-anchor quality would be refuted.

Watch

Extended reading notes

Core claim

On the paper's own terms, the central discovery is that the relative value of a text view is not a property of the prompt alone but depends on whether the visual encoder is frozen or adapted. Defining $\Delta_0 = A^0_D - A^0_B$ and $\Delta_L = A^L_D - A^L_B$ as Detailed-minus-Base accuracy before and after visual LoRA, the paper documents two recurring regimes. In semantic saturation, $\Delta_0 > 0$ but $0 < \Delta_L \ll \Delta_0$: the initial detailed-text advantage contracts from 8.13–21.54 percentage points to 0.69–2.96 points. In semantic emergence, $\Delta_0 \le 0$ but $\Delta_L > 0$: detailed descriptions become useful only after adaptation, by up to +3.84 points on ISIC. These regime assignments are supported by paired episode-level bootstrap intervals, training trajectories, sample-level transition statistics, shuffled-semantic controls, a second backbone, and multiple seeds; the paper's explicit conclusion is that frozen-model prompt quality is not a reliable proxy for adaptation-anchor quality.

Load-bearing premise

The load-bearing premise is that the class descriptions used as the Detailed text view were built only from a separate reference pool, verified by content hashes to be disjoint from every episode, so that the Detailed-minus-Base gap measures semantic utility rather than leaked access to target images.

Editorial extensions

If this is right

  • Prompt selection for SF-CDFSL should report $\Delta_0$, $\Delta_L$, and $\Delta_{\mathrm{shift}}$ on paired episodes, not just frozen zero-shot accuracy.
  • Reported zero-shot gains from detailed descriptions can overstate their post-adaptation value, since on EuroSAT and CropDisease most of the initial advantage is absorbed by Base-LoRA.
  • Prompts that look neutral or worse before adaptation can still be the better adaptation anchor, so discarding them on zero-shot evidence alone can sacrifice accuracy on domains like ISIC.
  • When language and supervision overlap, methods that assume semantic and adaptation gains add independently will double-count; methods that force preservation of frozen predictions may be counterproductive in emergence.
  • Visual LoRA is prompt-conditioned: cross-prompt evaluation shows nonzero transfer gaps, so the adapted model cannot be freely recombined with a different text coordinate system.

Reading between the lines

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

  • A practical extension the authors leave implicit: an automated prompt selector could watch the first few adaptation checkpoints, where query labels are not needed, and predict whether a description will saturate or emerge before committing to it.
  • If the mechanism is general, other adaptation families such as adapters, prompt tuning, or full fine-tuning may show the same sign flips, with the exact boundary depending on how quickly support supervision reshapes the visual space.
  • The protocol's dependence on an offline description bank means fair method comparisons must either give all baselines the same reference-image access or report the information-access difference alongside accuracy.
  • Across seeds and backbones, the stable object is the direction of utility shift rather than the categorical regime label, so future work should report continuous $\Delta_0$ and $\Delta_L$ instead of only naming regimes.
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. The paper investigates whether text-prompt rankings measured with a frozen vision–language model remain valid after source-free cross-domain few-shot adaptation with visual LoRA. Using a paired episodic protocol, it compares a generic class-name template (Base) with fixed detailed class descriptions (Detailed) on EuroSAT, CropDisease, ISIC, and ChestX, before and after adaptation. It defines zero-shot and adapted semantic utilities, Delta_0 and Delta_L, and reports two regimes: semantic saturation on EuroSAT/CropDisease, where a large initial Detailed advantage contracts after LoRA, and semantic emergence on ISIC/ChestX, where the Detailed view becomes superior only after adaptation. The paper supports these claims with training trajectories, a sample-level decomposition, a shuffled-semantic control, a second CLIP backbone, and additional seeds, and concludes that frozen zero-shot prompt quality is not a reliable proxy for adaptation-anchor quality.

Significance. If the empirical pattern is taken at face value, the paper makes a useful methodological point: prompt selection for source-free cross-domain few-shot learning should not be based solely on frozen-model accuracy. The paired episode design, bootstrap confidence intervals, sample-level algebraic decomposition, shuffled-semantic control, and multi-seed/multi-backbone checks are genuine strengths that raise the bar for empirical claims in this area. The central limitation is that the Detailed condition is constructed from target-domain reference images, so the Base–Detailed contrast conflates linguistic specificity with additional target-domain side information. This weakens the 'semantic utility' interpretation, although the underlying observation about adaptation-conditional utility remains empirically meaningful for the specific text views studied. The paper is honest about the limitation in its scope section, but the framing throughout, including the abstract and title, overstates the semantic nature of the effect.

major comments (3)
  1. [Appendix A.B and Table I] The main Base–Detailed comparison conflates linguistic specificity with information access. Appendix A.B states that each Detailed description was generated offline from one to three reference images drawn from the target datasets, and the Scope section concedes that Detailed introduces additional class-level target-domain side information. The content-hash verification only ensures that reference images do not appear in the reported episodes; it does not remove the structural asymmetry that the Detailed view receives target-domain visual statistics that the Base class-name view does not. Consequently, the Delta_0 and Delta_L gaps in Table I, and the saturation/emergence labels derived from them, can be driven by the amount of target-domain information encoded in the descriptions rather than by the semantic properties of the language alone. The shuffled-semantic control in Appendix E does not resolve this because it permutes the same possibly target-informed embeddings; it only establishes that class–description correspondence matters within that text bank. I recommend either reframing the central claim to describe the comparison between two text views that differ in both linguistic specificity and target-domain side information, or adding a control with descriptions generated without target reference images. The current term 'semantic utility' overstates what the design can establish.
  2. [Section V.B, Table III, and Table VIII] The ChestX 1-shot emergence result is too fragile to carry the weight the paper places on it. Table III reports Delta_L = +0.40 pp with a 95% confidence interval of [+0.03, +0.78] and p = 0.034, but the three-seed analysis in Table VIII shows mean Delta_L = +0.26 pp with standard deviation 0.29 pp, and the regime is not stable across seeds. The paper acknowledges this as a 'weak boundary case,' but the abstract and Section V.B still group ChestX with ISIC as a clean emergence example. Because the two-regime taxonomy is one of the paper's central contributions, ChestX 1-shot should either be assigned to a distinct 'weak emergence' category in the main text and abstract, or the emergence claim should be based on the more robust Delta_shift rather than on the positive sign of Delta_L. As written, the p-value of 0.034 conveys more stability than the seed analysis supports.
  3. [Appendix A.B and Appendix G] The exact description bank is not released, and the paper provides only the generation model and a high-level instruction. Since every empirical result in the paper depends on the specific wording and content of the Detailed prompts, independent verification is impossible without releasing the description bank, the reference-image identifiers, and the hash-based disjointness evidence. The contextual comparison in Appendix G is explicitly not harmonized, but the core paired comparison would be reproducible only if the frozen text bank is made public. I request that the authors release the full description bank and the audit artifacts used to verify the reference-pool disjointness.
minor comments (4)
  1. [Section IV.B.1] The phrase 'strictly paired protocol' should be qualified in the main text. Pairing holds for episode data, initialization, optimization, and evaluation, but not for information access, because the Detailed view is constructed with target-domain reference images. Appendix A.B explains this, but the main-text wording can be read as claiming a stronger form of control than is actually achieved.
  2. [Figure 2 and Appendix C] The endpoints of the 100-episode trajectory curves in Figure 2 differ from the principal endpoint estimates in Table I, and the text notes this only in the figure caption and appendix. A reader could misinterpret the trajectory endpoint as the main result. Adding a horizontal reference line or marker for the Table I endpoint in each panel would make the relationship clearer.
  3. [Appendix F] The prompt-transfer gaps in Table IX are large, especially on CropDisease (up to 22.22 pp for Detailed-trained LoRA evaluated with the Base prompt). The paper explains that positive gaps favor the training prompt, but it does not offer a mechanistic explanation for why switching prompts after training causes such a large drop. A brief discussion of this asymmetry would help readers interpret the co-adaptation claim.
  4. [Appendix E, Table VII] The ISIC 1-shot ViT-B/32 result is labeled AMPL (amplification) because Delta_0 is slightly positive, but no confidence interval or p-value is provided for that cell. Since the regime label changes relative to ViT-B/16, reporting the bootstrap uncertainty for this cell would make the re-labeling more transparent.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: the paper's central claim is an empirical measurement, not a result forced by its definitions or by self-citation.

full rationale

The paper's central quantities, Delta_0 and Delta_L, are directly measured accuracy differences under a strictly paired episodic protocol, and the saturation/emergence regimes are operational labels assigned from the observed signs of these measured quantities (Eqs. 14-19). No parameter is fitted to the target claim, and no 'prediction' is derived from an input that already contains the output. The sample-level decomposition in Eq. (28) is an explicit algebraic identity, as the appendix demonstrates by set partitioning; the paper does not present it as a fitted or predicted result. The Detailed description bank is an information-access difference relative to the Base class-name condition, and this is disclosed and audited in Appendix A.B; while this affects the interpretation of the Base-Detailed contrast as a measure of purely linguistic utility, it is a validity concern rather than circularity, because the paper's conclusion does not presuppose the sign of the measured utility difference. The shuffled-semantic control, second backbone, and multi-seed runs are external checks, and the paper explicitly flags ChestX 1-shot as a weak boundary case. Self-citation is not load-bearing: the cited prior work is by other research groups, and no uniqueness theorem or prior author result is invoked to force the chosen interpretation. The central claim that frozen-model prompt quality is an incomplete proxy for adaptation-anchor quality follows from the observed mismatch between zero-shot and post-adaptation rankings, which is an empirical finding rather than a tautology.

Assumptions & free parameters 8 free parameters · 6 assumptions · 0 invented entities

The central claim rests on a controlled empirical protocol rather than a mathematical derivation. The only inputs pulled from prior work are CLIP and LoRA; the core free choices are the fixed adaptation hyperparameters and the LLM-generated description bank. No entity is invented; the regime labels are descriptive.

free parameters (8)
  • LoRA rank = 16
    Chosen rank for visual adaptation; fixed and identical for Base and Detailed conditions.
  • LoRA scaling factor = 1
    Fixed scaling of the low-rank update, identical for both text views.
  • LoRA dropout = 0.25
    Dropout applied to LoRA factors, identical across conditions.
  • AdamW learning rate and weight decay = lr=1e-4, wd=1e-2
    Optimizer settings fixed before analysis and shared by Base and Detailed runs.
  • Optimizer updates per episode = 250 (ChestX 125)
    Dataset-specific budget fixed before analysis and identical for Base/Detailed, preserving paired comparison.
  • Logit scale = 100
    Cosine similarity temperature, fixed for all evaluations.
  • Description generation model and instruction = Qwen3.5-27B-FP8 with instruction restricting to observable visual evidence
    The exact LLM and prompting instruction are a hand-chosen text-generation setup; the generated bank is frozen and reused.
  • Reference images per class = 1-3
    Number of target-domain reference images used to generate descriptions; this is class-level side information.
assumptions (6)
  • standard math CLIP visual and text encoders provide a meaningful zero-shot classifier under cosine similarity with logit scale 100
    Background assumption for the entire evaluation, invoked in Section III.
  • domain assumption Low-rank adaptation of query, key, and value projections in all ViT blocks is a valid controlled adaptation probe
    The paper's conclusion is conditional on visual LoRA; Section IV.B fixes this configuration.
  • domain assumption The paired episode protocol (same classes, support, query, initialization, random streams) isolates the effect of the text view
    Section IV.A and Appendix A state this separation; it is the logical basis for Delta 0 / Delta L comparisons.
  • domain assumption The Detailed description bank is class-level target-domain side information generated from reference images disjoint from the episodes
    Appendix A.B; if this separation failed, the comparison would leak target information into the Detailed condition.
  • domain assumption The fixed-point-free shuffled control isolates semantic correspondence from text-vector geometry
    Section VIII and Appendix E; the control preserves the embedding multiset while breaking class-description pairing.
  • standard math Bootstrap percentile intervals with episode-level resampling are valid for the paired differences
    Appendix A.D, following Efron and Tibshirani [49].

how reviews work

0 comments
Cite this review

Pith. "Pith review of When Semantics Saturate or Emerge: Adaptation-Conditional Semantic Utility in Source-Free Cross-Domain Few-Shot Learning." pith.science (2026). https://pith.science/paper/NPHXHJD2

@misc{pith2026260806673,
  author       = {Pith},
  title        = {Pith review of: When Semantics Saturate or Emerge: Adaptation-Conditional Semantic Utility in Source-Free Cross-Domain Few-Shot Learning},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/NPHXHJD2}},
  note         = {Machine review of arXiv:2608.06673}
}
abstract

Language descriptions in source-free cross-domain few-shot learning (SF-CDFSL) are often selected according to zero-shot accuracy obtained with a frozen vision--language model. This paper asks whether that ranking remains valid after target-domain visual adaptation. Under a strictly paired protocol, we compare a generic class-name template with fixed detailed class descriptions before and after visual Low-Rank Adaptation (LoRA) on EuroSAT, CropDisease, ISIC, and ChestX. Let $\deltazero$ and $\deltalora$ denote the Detailed-minus-Base accuracy before and after adaptation, respectively. Two recurring regimes emerge. In \emph{semantic saturation}, $\deltazero>0$ but $0<\deltalora\ll\deltazero$: on EuroSAT and CropDisease, initial gains of 8.13--21.54 percentage points contract to 0.69--2.96 points after LoRA. In \emph{semantic emergence}, $\deltazero\leq0$ but $\deltalora>0$: on ISIC and ChestX, detailed descriptions become more useful only after the visual representation is updated. Training trajectories and sample-level decomposition show that saturation is driven mainly by Base-LoRA recovering errors already solved by detailed semantics, whereas emergence is associated with prediction turnover and newly formed Detailed-only correct decisions. Fixed-point-free shuffled-semantic controls, a second CLIP backbone, and multiple random seeds support the broad pattern while identifying ChestX 1-shot as a weak boundary case. These findings establish that zero-shot prompt quality is an incomplete proxy for adaptation-anchor quality and motivate evaluating language on both sides of the adaptation boundary.

Figures

Figures reproduced from arXiv: 2608.06673 by the authors.

Figure 1
Figure 1. Adaptation-conditional semantic utility. The horizontal axis measures Detailed-minus-Base utility before visual LoRA ( [PITH_FULL_IMAGE:figures/full_fig_p006_1.png] view at source ↗
Figure 2
Figure 2. Evolution of matched-view semantic utility during visual LoRA adaptation. At checkpoint [PITH_FULL_IMAGE:figures/full_fig_p007_2.png] view at source ↗
Figure 3
Figure 3. Sample-level evidence for the two semantic-utility regimes. Detailed-ZS retention measures the fraction of frozen Detailed successes preserved by [PITH_FULL_IMAGE:figures/full_fig_p009_3.png] view at source ↗
Figures from the paper (3 more)
Figure 4
Figure 4. Figure 4: Aligned Detailed accuracy minus a fixed-point-free shuffled Detailed codebook. The control preserves the text-embedding multiset but breaks class– [PITH_FULL_IMAGE:figures/full_fig_p010_4.png]
Figure 5
Figure 5. Figure 5: Per-domain training trajectories. EuroSAT and CropDisease exhibit delayed contraction after an initially stable or amplified Detailed advantage. ISIC [PITH_FULL_IMAGE:figures/full_fig_p014_5.png]
Figure 6
Figure 6. Figure 6: Prompt-transfer gaps after Base-conditioned and Detailed-conditioned [PITH_FULL_IMAGE:figures/full_fig_p017_6.png]

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

49 extracted references · 36 canonical work pages

  1. [1]

    A broader study of cross-domain few-shot learning,

    Y . Guo, N. C. F. Codella, L. Karlinsky, J. V . Codella, J. R. Smith, K. Saenko, T. Rosing, and R. Feris, “A broader study of cross-domain few-shot learning,” inEuropean Conference on Computer Vision, 2020, pp. 124–141

  2. [2]

    Visual domain bridge: A source-free domain adaptation for cross-domain few-shot learning,

    M. Yazdanpanah and P. Moradi, “Visual domain bridge: A source-free domain adaptation for cross-domain few-shot learning,” inProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recogni- tion Workshops, 2022, pp. 2868–2877

  3. [3]

    Enhanc- ing information maximization with distance-aware contrastive learning for source-free cross-domain few-shot learning,

    H. Xu, L. Liu, S. Zhi, S. Fu, Z. Su, M.-M. Cheng, and Y . Liu, “Enhanc- ing information maximization with distance-aware contrastive learning for source-free cross-domain few-shot learning,”IEEE Transactions on Image Processing, vol. 33, pp. 2058–2073, 2024

  4. [4]

    Learning transferable visual models from natural language supervision,

    A. Radford, J. W. Kim, C. Hallacy, A. Ramesh, G. Goh, S. Agarwal, G. Sastry, A. Askell, P. Mishkin, J. Clarket al., “Learning transferable visual models from natural language supervision,” inInternational Conference on Machine Learning, 2021, pp. 8748–8763

  5. [5]

    Learning to prompt for vision- language models,

    K. Zhou, J. Yang, C. C. Loy, and Z. Liu, “Learning to prompt for vision- language models,”International Journal of Computer Vision, vol. 130, no. 9, pp. 2337–2348, 2022

  6. [6]

    MaPLe: Multi-modal prompt learning,

    M. U. Khattak, H. Rasheed, M. Maaz, S. Khan, and F. S. Khan, “MaPLe: Multi-modal prompt learning,” inProceedings of the IEEE/CVF Con- ference on Computer Vision and Pattern Recognition, 2023, pp. 19 113– 19 122

  7. [7]

    LoRA: Low-rank adaptation of large language models,

    E. J. Hu, Y . Shen, P. Wallis, Z. Allen-Zhu, Y . Li, S. Wang, L. Wang, and W. Chen, “LoRA: Low-rank adaptation of large language models,” inInternational Conference on Learning Representations, 2022

  8. [8]

    Low-rank few-shot adaptation of vision- language models,

    M. Zanella and I. Ben Ayed, “Low-rank few-shot adaptation of vision- language models,” inProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition Workshops, 2024, pp. 1593– 1603

Show all 49 references
  1. [9]

    Prompt as free lunch: En- hancing diversity in source-free cross-domain few-shot learning through semantic-guided prompting,

    L. Zhuo, Z. Wang, Y . Fu, and T. Qian, “Prompt as free lunch: En- hancing diversity in source-free cross-domain few-shot learning through semantic-guided prompting,”arXiv preprint arXiv:2412.00767, 2024

  2. [10]

    Step-wise distribution-aligned style prompt tuning for source-free cross-domain few-shot learning,

    H. Xu, L. Liu, T. Liu, S. Zhi, S. Sun, and M.-M. Cheng, “Step-wise distribution-aligned style prompt tuning for source-free cross-domain few-shot learning,”IEEE Transactions on Pattern Analysis and Machine Intelligence, vol. 48, no. 2, pp. 1098–1114, 2026. 11

  3. [11]

    Addressing exacerbated attention sink for source-free cross-domain few-shot learning,

    S. Yi, Y . Zou, Y . Li, and R. Li, “Addressing exacerbated attention sink for source-free cross-domain few-shot learning,” inProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, 2026, pp. 29 494–29 503

  4. [12]

    Reviving in-domain fine- tuning methods for source-free cross-domain few-shot learning,

    Y . Zhao, Y . Liu, Y . Zou, Y . Li, and R. Li, “Reviving in-domain fine- tuning methods for source-free cross-domain few-shot learning,”arXiv preprint arXiv:2605.11659, 2026

  5. [13]

    Mind the discriminability trap in source-free cross-domain few-shot learning,

    Z. Zhang, Y . Zou, Y . Li, R. Li, and G. Chen, “Mind the discriminability trap in source-free cross-domain few-shot learning,” inProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, 2026, pp. 36 978–36 988

  6. [14]

    Reclaiming lost text layers for source-free cross-domain few-shot learning,

    Z. Zhang, G. Chen, Y . Zou, Y . Li, and R. Li, “Reclaiming lost text layers for source-free cross-domain few-shot learning,” inProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, 2026, pp. 15 080–15 090

  7. [15]

    Interpretable cross-domain few-shot learning with rectified target-domain local alignment,

    Y . Zhao, Y . Zou, Y . Li, and R. Li, “Interpretable cross-domain few-shot learning with rectified target-domain local alignment,” inProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, 2026, pp. 41 605–41 615

  8. [16]

    Matching networks for one shot learning,

    O. Vinyals, C. Blundell, T. Lillicrap, K. Kavukcuoglu, and D. Wierstra, “Matching networks for one shot learning,” inAdvances in Neural Information Processing Systems, 2016, pp. 3637–3645

  9. [17]

    Model-agnostic meta-learning for fast adaptation of deep networks,

    C. Finn, P. Abbeel, and S. Levine, “Model-agnostic meta-learning for fast adaptation of deep networks,” inInternational Conference on Machine Learning, 2017, pp. 1126–1135

  10. [18]

    Prototypical networks for few-shot learning,

    J. Snell, K. Swersky, and R. Zemel, “Prototypical networks for few-shot learning,” inAdvances in Neural Information Processing Systems, 2017, pp. 4077–4087

  11. [19]

    Learning to compare: Relation network for few-shot learning,

    F. Sung, Y . Yang, L. Zhang, T. Xiang, P. H. S. Torr, and T. M. Hospedales, “Learning to compare: Relation network for few-shot learning,” inProceedings of the IEEE Conference on Computer Vision and Pattern Recognition, 2018, pp. 1199–1208

  12. [20]

    A closer look at few-shot classification,

    W.-Y . Chen, Y .-C. Liu, Z. Kira, Y .-C. F. Wang, and J.-B. Huang, “A closer look at few-shot classification,” inInternational Conference on Learning Representations, 2019

  13. [21]

    Meta-dataset: A dataset of datasets for learning to learn from few examples,

    E. Triantafillou, T. Zhu, V . Dumoulin, P. Lamblin, U. Evci, K. Xu, R. Goroshin, C. Gelada, K. Swersky, P.-A. Manzagol, and H. Larochelle, “Meta-dataset: A dataset of datasets for learning to learn from few examples,” inInternational Conference on Learning Representations, 2020

  14. [22]

    Cross-domain few-shot classification via learned feature-wise transformation,

    H.-Y . Tseng, H.-Y . Lee, J.-B. Huang, and M.-H. Yang, “Cross-domain few-shot classification via learned feature-wise transformation,” inIn- ternational Conference on Learning Representations, 2020

  15. [23]

    Understand- ing cross-domain few-shot learning based on domain similarity and few- shot difficulty,

    J. Oh, S. Kim, N. Ho, J.-H. Kim, H. Song, and S.-Y . Yun, “Understand- ing cross-domain few-shot learning based on domain similarity and few- shot difficulty,” inAdvances in Neural Information Processing Systems, vol. 35, 2022

  16. [24]

    Cross-domain few-shot learning with task-specific adapters,

    W.-H. Li, X. Liu, and H. Bilen, “Cross-domain few-shot learning with task-specific adapters,” inProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, 2022, pp. 7161–7170

  17. [25]

    Ranking distance calibration for cross-domain few-shot learning,

    P. Li, S. Gong, C. Wang, and Y . Fu, “Ranking distance calibration for cross-domain few-shot learning,” inProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, 2022, pp. 9099–9108

  18. [26]

    Revisiting proto- typical network for cross domain few-shot learning,

    F. Zhou, P. Wang, L. Zhang, W. Wei, and Y . Zhang, “Revisiting proto- typical network for cross domain few-shot learning,” inProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, 2023, pp. 20 061–20 070

  19. [27]

    Wave-SAN: Wavelet based style augmentation network for cross-domain few-shot learning,

    Y . Fu, Y . Xie, Y . Fu, J. Chen, and Y .-G. Jiang, “Wave-SAN: Wavelet based style augmentation network for cross-domain few-shot learning,” inProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, 2022

  20. [28]

    StyleAdv: Meta style adversarial training for cross-domain few-shot learning,

    Y . Fu, Y . Xie, Y . Fu, and Y .-G. Jiang, “StyleAdv: Meta style adversarial training for cross-domain few-shot learning,” inProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, 2023, pp. 24 575–24 584

  21. [29]

    Flatten long-range loss landscapes for cross-domain few-shot learning,

    Y . Zou, Y . Liu, Y . Hu, Y . Li, and R. Li, “Flatten long-range loss landscapes for cross-domain few-shot learning,” inProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, 2024, pp. 23 575–23 584

  22. [30]

    Reconstruction target matters in masked image modeling for cross-domain few-shot learning,

    R. Ma, Y . Zou, Y . Li, and R. Li, “Reconstruction target matters in masked image modeling for cross-domain few-shot learning,” inProceedings of the AAAI Conference on Artificial Intelligence, 2025, pp. 19 305–19 313

  23. [31]

    Attention temperature matters in ViT-based cross-domain few-shot learning,

    Y . Zou, R. Ma, Y . Li, and R. Li, “Attention temperature matters in ViT-based cross-domain few-shot learning,” inAdvances in Neural Information Processing Systems, vol. 37, 2024, pp. 116 332–116 354

  24. [32]

    A closer look at the CLS token for cross-domain few-shot learning,

    Y . Zou, S. Yi, Y . Li, and R. Li, “A closer look at the CLS token for cross-domain few-shot learning,” inAdvances in Neural Information Processing Systems, vol. 37, 2024, pp. 85 523–85 545

  25. [33]

    Revisiting continuity of image tokens for cross-domain few-shot learning,

    S. Yi, Y . Zou, Y . Li, and R. Li, “Revisiting continuity of image tokens for cross-domain few-shot learning,”arXiv preprint arXiv:2506.03110, 2025

  26. [34]

    Random registers for cross-domain few-shot learning,

    ——, “Random registers for cross-domain few-shot learning,”arXiv preprint arXiv:2506.02843, 2025

  27. [35]

    Conditional prompt learning for vision-language models,

    K. Zhou, J. Yang, C. C. Loy, and Z. Liu, “Conditional prompt learning for vision-language models,” inProceedings of the IEEE/CVF Confer- ence on Computer Vision and Pattern Recognition, 2022, pp. 16 816– 16 825

  28. [36]

    Visual-language prompt tuning with knowledge-guided context optimization,

    H. Yao, R. Zhang, and C. Xu, “Visual-language prompt tuning with knowledge-guided context optimization,” inProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, 2023, pp. 6757–6767

  29. [37]

    Prompt-aligned gradient for prompt tuning,

    B. Zhu, Y . Niu, Y . Han, Y . Wu, and H. Zhang, “Prompt-aligned gradient for prompt tuning,” inProceedings of the IEEE/CVF International Conference on Computer Vision, 2023, pp. 15 659–15 669

  30. [38]

    Self-regulating prompts: Foundational model adaptation without forgetting,

    M. U. Khattak, S. T. Wasim, M. Naseer, S. Khan, M.-H. Yang, and F. S. Khan, “Self-regulating prompts: Foundational model adaptation without forgetting,” inProceedings of the IEEE/CVF International Conference on Computer Vision, 2023, pp. 15 190–15 200

  31. [39]

    Tip-adapter: Training-free adaption of CLIP for few-shot classification,

    R. Zhang, R. Zhang, R. Fang, P. Gao, K. Li, J. Dai, Y . Qiao, and H. Li, “Tip-adapter: Training-free adaption of CLIP for few-shot classification,” inEuropean Conference on Computer Vision, 2022, pp. 493–510

  32. [40]

    CLIP-adapter: Better vision-language models with feature adapters,

    P. Gao, S. Geng, R. Zhang, T. Ma, R. Fang, Y . Zhang, H. Li, and Y . Qiao, “CLIP-adapter: Better vision-language models with feature adapters,” International Journal of Computer Vision, vol. 132, no. 2, pp. 581–595, 2024

  33. [41]

    Task residual for tuning vision-language models,

    T. Yu, Z. Lu, X. Jin, Z. Chen, and X. Wang, “Task residual for tuning vision-language models,” inProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, 2023, pp. 10 899–10 909

  34. [42]

    What does a platypus look like? generating customized prompts for zero-shot image classification,

    S. Pratt, I. Covert, R. Liu, and A. Farhadi, “What does a platypus look like? generating customized prompts for zero-shot image classification,” inProceedings of the IEEE/CVF International Conference on Computer Vision, 2023, pp. 15 691–15 701

  35. [43]

    Visual classification via description from large language models,

    S. Menon and C. V ondrick, “Visual classification via description from large language models,” inInternational Conference on Learning Rep- resentations, 2023

  36. [44]

    Improving CLIP adaptation by breaking tail alignment for source-free cross-domain few-shot learning,

    S. Yi, Y . Zou, Y . Li, and R. Li, “Improving CLIP adaptation by breaking tail alignment for source-free cross-domain few-shot learning,” inInternational Conference on Machine Learning, ser. Proceedings of Machine Learning Research, vol. 306, 2026

  37. [45]

    EuroSAT: A novel dataset and deep learning benchmark for land use and land cover classi- fication,

    P. Helber, B. Bischke, A. Dengel, and D. Borth, “EuroSAT: A novel dataset and deep learning benchmark for land use and land cover classi- fication,”IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing, vol. 12, no. 7, pp. 2217–2226, 2019

  38. [46]

    Using deep learning for image-based plant disease detection,

    S. P. Mohanty, D. P. Hughes, and M. Salath ´e, “Using deep learning for image-based plant disease detection,”Frontiers in Plant Science, vol. 7, p. 1419, 2016

  39. [47]

    Skin lesion analysis toward melanoma detection 2018: A challenge hosted by the international skin imaging collaboration,

    N. C. F. Codella, V . Rotemberg, P. Tschandl, M. E. Celebi, S. Dusza, D. Gutman, B. Helba, A. Kalloo, K. Liopyris, M. Marchettiet al., “Skin lesion analysis toward melanoma detection 2018: A challenge hosted by the international skin imaging collaboration,”arXiv preprint arXiv...

  40. [48]

    ChestX-ray8: Hospital-scale chest x-ray database and benchmarks on weakly-supervised classification and localization of common thorax diseases,

    X. Wang, Y . Peng, L. Lu, Z. Lu, M. Bagheri, and R. M. Summers, “ChestX-ray8: Hospital-scale chest x-ray database and benchmarks on weakly-supervised classification and localization of common thorax diseases,” inProceedings of the IEEE Conference on Computer Vision and Pattern...

  41. [49]

    practical significance

    B. Efron and R. J. Tibshirani,An Introduction to the Bootstrap. Boca Raton, FL: CRC Press, 1994. 12 APPENDIXA REPRODUCIBILITY ANDINFORMATIONACCESS A. Target Domains and Episode Construction The four target datasets follow the BSCD-FSL evaluation family: EuroSAT for remote-sens...

Pith tools

Reviewed August 10, 2026 · model on record in the stance chip above.