Pith. sign in

REVIEW 4 major objections 5 minor 41 references

DoubleCCA: Improving Foundation Model Group Robustness with Random Sentence Embeddings

T0 review · 4 major / 5 minor · reviewed 2026-08-12 · deepseek-v4-flash

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

Pith's one-line read The paper claims that a text-only, training-free CCA-based merge of CLIP and sentence-embedding representations improves worst-group zero-shot accuracy, with Waterbirds ResNet-50 worst-group accuracy rising from 16.07% to 44.64%.

desk verdict A clever, training-free debiasing recipe combining WaffleCLIP prompts with CCA merging, but the empirical support is under-powered and the second CCA's geometry is underspecified. read the letter →

arxiv 2411.16236 v1 pith:KP3GTUON submitted 2024-11-25 cs.CL cs.CV

Hong Liu , Yitong Lu This is my paper · ORCID
classification cs.CLcs.CV
keywords grouprobustnessCLIPzero-shotclassificationcanonicalcorrelationanalysissentenceembeddingsrandompromptaugmentationspuriouscorrelationstext-onlydebiasing
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 proposes DoubleCCA, a training-free recipe for making CLIP zero-shot classifiers more robust to group-based biases, such as a background that correlates with the label. The idea is to augment each class prompt with many random sentences, embed them with both the CLIP text encoder and a separate sentence embedding model, and then align and merge the two embedding spaces with two passes of canonical correlation analysis (CCA). The result is a single merged text-embedding matrix that replaces the original prompt embeddings at inference time. The paper reports that on Waterbirds with a ResNet-50 backbone, worst-group accuracy rises from 16.07% to 44.64% with ordinary prompts and from 39.29% to 48.21% when background context is included, with smaller gains on CelebA and with domain-generalization accuracy roughly maintained across six datasets. If correct, this would mean group robustness can be improved with text alone, without image data, group labels, or fine-tuning.

What carries the argument

The central object is the merged text-embedding matrix $W = \tfrac{1}{2}(\hat{W}_x + M\hat{W}_{se})$, where $\hat{W}_x$ and $\hat{W}_{se}$ are the reconstructed score-function weights from the first CCA and $M = (P_B P_A^{-1})^T$ is the merger learned by the second CCA. Canonical correlation analysis (CCA) is the statistical method that finds linear projections maximizing correlation between two feature sets; here it is applied twice. The first pass aligns CLIP text embeddings with sentence-embedding-model outputs on a corpus of random sentences, and the second pass merges the two score functions using random-sentence features as stand-ins for image embeddings, since no images are available. The machinery carries the argument by turning an uncalibrated random-prompt augmentation into a calibrated linear merge that plugs directly into the existing CLIP zero-shot score $\hat{y} = \arg\max_y W_y \Phi_v(I)$.

What would settle it

Run DoubleCCA with the second alignment fitted on real image embeddings from a held-out split (or on dimension-matched random noise) instead of random-sentence embeddings and compare worst-group accuracy: if the gains disappear or change materially, the random-sentence proxy is doing the work; if not, the claimed alignment mechanism is not the cause.

Watch

Extended reading notes

Core claim

The central claim is that canonical correlation analysis can turn noisy random-prompt text embeddings into a more group-robust zero-shot classifier by merging two text-embedding sources. Specifically, the paper constructs the merged text embedding matrix $$W = \tfrac{1}{2}(\hat{W}_x + M\hat{W}_{se}), \qquad M = (P_B $P_A^{{-1}}$)^T,$$ where the first CCA aligns the CLIP text encoder with a sentence embedding model on random-sentence features, and the second CCA merges the two resulting score functions into one. Inference is then unchanged in form: $\hat{y} = \arg\max_y W_y \Phi_v(I)$. The paper argues that this merged embedding improves worst-group accuracy because the extra sentence encoder supplies semantically meaningful structure that the CLIP text encoder lacks, while the CCA alignment prevents the random tokens from collapsing class separability, as WaffleCLIP does on smaller backbones. The empirical case is made on Waterbirds and CelebA across four CLIP backbones, plus six domain-generalization datasets.

Load-bearing premise

The recipe's gains rest on the assumption that text embeddings of random sentences behave like real image embeddings during the second alignment step, since the method never looks at an actual image.

Editorial extensions

If this is right

  • If the central claim holds, zero-shot CLIP classification can be made more group-robust without any image data, group annotations, or gradient updates, simply by substituting $W$ for the original text embeddings.
  • Because $W$ is built only from class names and random sentences, the recipe transfers across backbones (RN50, ViT-B/32, ViT-B/16, ViT-L/14) and datasets, improving worst-group accuracy on Waterbirds and CelebA while keeping domain-generalization accuracy roughly constant.
  • The merged embeddings can be dropped into existing adapter-based debiasing pipelines: combining DoubleCCA with a contrastive adapter further lifts worst-group accuracy, so the text-only enrichment is complementary rather than exclusive.
  • Sufficient random sentences (about 500 per class) are required for stable worst-group gains; with fewer random sentences the method can fall below the CLIP baseline, implying that the text-only augmentation distribution, not the class descriptions alone, carries the robustness signal.

Reading between the lines

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

  • A direct test of the paper's mechanism: fit the second CCA on a held-out set of real image embeddings instead of random-sentence proxies. If the worst-group gains vanish, the proxy assumption is doing the work; if they persist, the paper's stated role of random sentences as image stand-ins is not the active ingredient.
  • Because the merged embedding is computed entirely from text, the recipe suggests a path to debiasing open-vocabulary or newly added classes on the fly, as long as a sentence encoder is available; this extension is not tested in the paper.
  • The ablation curve with sentence count implies the random-sentence distribution is the carrier of robustness. One could therefore try structured or semantic perturbations of the random sentences (e.g., class-related nouns instead of characters) and expect larger worst-group gains, though the paper does not explore this.
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

4 major / 5 minor

Summary. The paper proposes DoubleCCA, a training-free method to improve the worst-group robustness of CLIP zero-shot classifiers. The method generates K random sentences per class, embeds them with both the CLIP text encoder and an external sentence embedding model (HiT), and then applies CCA twice: first to align the two text embedding spaces, and second to merge the two resulting score functions into a single text embedding matrix. At inference, this merged matrix is used to score CLIP image embeddings. Experiments on Waterbirds and CelebA report improved worst-group accuracy across four backbones, and results on six domain-generalization datasets show maintained or slightly improved accuracy. The central claim is that random-sentence text enrichment plus CCA merging improves group robustness without using image data, group labels, or training.

Significance. If the central claim holds, the method is an interesting training-free debiasing recipe for CLIP: it requires no image data or group labels, is conceptually simple, and can be combined with existing adapters. The paper includes useful ablations on the sentence embedding model and on hyperparameters, and it addresses a relevant problem. However, the evidence is currently weak: no code, no seeds, and no error bars are reported; the Waterbirds worst-group test set is very small; and there is a load-bearing dimension inconsistency in the second CCA construction. The assumption that random-sentence text features can simulate image representations is asserted but not validated, leaving the mechanism unsupported. These issues must be resolved before the reported robustness gains can be taken as established.

major comments (4)
  1. [Section 3.2.2 / Algorithm 1 step 7] The construction XA = Ŵx Fr, XB = Ŵse Fr is dimensionally inconsistent with the stated experimental settings. For Waterbirds with RN50, n_classes=2, d=1024, and K=500, Fr has 1000 rows. If Fr is n_random×d, the products ŴxFr and ŴseFr are undefined because Ŵx and Ŵse are 2×1024; if Fr is d×n_random, then XA and XB are 2×1000, so after centering the second CCA has rank at most 1 and cannot produce a common space of dimension equal to the original image embedding dimension (1024), as stated in Section 4.1. The paper must specify the exact shapes of Fr, XA, XB, the centering and regularization used in the second CCA, and ideally release code; otherwise the reported numbers cannot be reproduced or attributed to the described mechanism.
  2. [Section 3.2.2] The statement that random sentence embedding features Fr 'simulate the image representations' is the sole bridge between text-only calibration and image inference, but it is asserted without evidence. The second CCA is fit on XA = Ŵx Fr and XB = Ŵse Fr, i.e., on the text-based score functions evaluated on random sentences, while at test time the merged matrix W is applied to real image embeddings. This is a domain shift that the paper does not validate. To support the central claim, the authors should provide quantitative evidence that Fr-derived features are representative of image features in the relevant space, for example by comparing distributions, or by ablating the second CCA with real image features from a validation set. Without such validation, the reported worst-group improvements cannot be confidently assigned to the proposed calibration mechanism.
  3. [Section 4.1 / Table 1] No seeds, error bars, or confidence intervals are reported anywhere. The Waterbirds worst-group test split is small (about 56 images), so the headline RN50 improvement from 16.07% to 44.64% corresponds to a change of roughly 9 to 25 correct images, which could arise from sampling noise. Please report means and standard deviations over at least 5 random seeds for all reported numbers, and state the exact sizes of the worst-group test sets for Waterbirds and CelebA. The same applies to the domain generalization results in Table 2.
  4. [Section 4.2 / Table 1] The text compares DoubleCCA with PerceptionCLIP and FairerCLIP, but these baselines do not appear in Table 1, and the reported metric 'gap' is never defined (presumably average accuracy minus worst-group accuracy). Please include the baselines in the table or explicitly state that the comparison is qualitative, and define 'gap' when it is first used. Without the baselines, the claims of outperforming existing methods are not directly supported by the presented experiments.
minor comments (5)
  1. [Section 3.2.1 and 3.2.2] The symbol Fr is reused in two different roles: first as the random-sentence embedding matrix for the first CCA, and then as the proxy for image representations in the second CCA. This conflates two distinct objects and contributes to the dimension ambiguity; please use different names, e.g., F_r^{text} for the first CCA and F_r^{proxy} for the second.
  2. [Abstract] The phrase 'use CCA double twice' should be corrected to 'use CCA twice'.
  3. [Section 4.2] There are several typos: 'backgroup' should be 'background', 'folowers' should be 'flowers', and 'with folowers' appears in the list of background contexts. The heading in Section 4.5, 'Abalation Study', should be 'Ablation Study'.
  4. [Figure 6] In the caption of Figure 6(b), 'Dimension of CCA' is ambiguous: Section 4.1 sets the first CCA dimension to 64, while the ablation varies the dimension up to 384. Please specify which CCA (first or second) is being varied, and clarify whether the second CCA dimension is also fixed or follows the stated rule.
  5. [Section 4.1] The paper says it follows 'the same experimental settings as the previous work [11]' but does not specify which elements are taken from that work (prompt templates, data splits, evaluation protocol). Please list the exact settings used, including the prompt template and the source of the random sentences, to make the experiments reproducible.

Circularity Check

0 steps flagged · score 0.0 of 10

No circularity: the merged embedding is constructed from text-only features and evaluated against real images, so the reported gains are not forced by the fitting equations.

full rationale

The derivation chain is not circular: the final merged embedding W in Eq. (5) is a deterministic function of the original class-prompt embeddings and the random-sentence embeddings Fr/Frse, and the second CCA in Eqs. (6)-(7) is fit on XA = Wx Fr and XB = Wse Fr, i.e., entirely on text features. No group label, image feature, or worst-group accuracy enters the construction of W; the inference rule in Eq. (8) then applies W to real image features, so the reported worst-group improvements are an external empirical consequence rather than a consequence of the fitting procedure. The stated assumption that Fr 'simulate the image representations' (Sect. 3.2.2) is an unvalidated proxy and a reproducibility risk, and the dimensional inconsistency noted in Sect. 4.1 is a correctness concern, but neither reduces the central claim to its inputs by construction. There are no load-bearing self-citations or imported uniqueness theorems. Accordingly, no significant circularity is found.

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

Free parameters: six design choices, of which five are set by hand or by ablation on the Waterbirds benchmark rather than by theory, and one (CCA regularization) is never reported. Axioms: the method assumes random-sentence CLIP text embeddings can stand in for images when fitting the merge, that HiT features add useful semantic structure, that CCA transforms fit on random sentences transfer to original prompts and real images, and that the merge rule of [23] preserves discriminative power. Invented entities: none. The ledger makes clear that the load-bearing machinery is imported from [10], [21], and [23], with the paper's contribution being the specific combination and the proxy trick.

free parameters (6)
  • Number of random sentences K = 500
    Set to 500 per class in Section 4.1; Section 4.5.1 shows worst-group robustness is highly variable for K below 500 and stabilizes at larger K; selected by Waterbirds ablation, not by theory.
  • First CCA dimension = 64
    Section 4.5.1 ablation on Waterbirds with ResNet-50; both lower and higher dimensions hurt accuracy, so 64 is chosen as best on the evaluation benchmark.
  • CCA regularization strength = unspecified
    Section 2 states 'we use regularized CCA [22, 23]' but no regularization value is reported anywhere; this affects both CCA steps and blocks exact reproduction.
  • Random character sequence length = length of the class name
    Adopted from WaffleCLIP [10, 11] per Section 4.1; this design choice shapes the text embedding statistics used by both CCA passes.
  • Sentence embedding model choice = HiT-MiniLM-L12-WordNetNoun
    Selected because it shows the most significant improvements in the Waterbirds ablation of Section 4.4; BART, which performs worse, is rejected.
  • Second CCA dimension = CLIP embedding dimension d
    Set to the original image embedding dimension as the reconstruction target in Section 4.1; a fixed design choice rather than a fitted value.
assumptions (4)
  • ad hoc to paper CLIP text embeddings of random sentences can stand in for image embeddings when fitting the second CCA merge
    Invoked in Section 3.2.2 ('we further utilize the random sentence embedding features Fr that simulate the image representations') and Algorithm 1 steps 7 to 9; the text-to-image distribution mismatch is never quantified.
  • domain assumption The sentence embedding model (HiT, trained on WordNet) produces semantically meaningful features that complement CLIP text features for zero-shot classification
    Assumed in Section 3.2 and supported only by the Waterbirds ablation in Section 4.4; no analysis of what structure HiT adds beyond generic sentence embeddings.
  • domain assumption CCA transforms fit on random-sentence rows transfer to the original class prompts and to real image embeddings
    Both CCA steps are fit on random-sentence features, while the final evaluation uses original prompt embeddings and real image embeddings; transfer of the transforms is assumed, not demonstrated.
  • standard math The merge rule W = 1/2(Ŵx + M·Ŵse) from Horoi et al. preserves enough discriminative information for zero-shot classification
    Borrowed from reference [23] in Eq. 5; validity in this text-embedding setting is inherited from that prior work rather than re-derived here.

how reviews work

0 comments
Cite this review

Pith. "Pith review of DoubleCCA: Improving Foundation Model Group Robustness with Random Sentence Embeddings." pith.science (2026). https://pith.science/paper/KP3GTUON

@misc{pith2026241116236,
  author       = {Pith},
  title        = {Pith review of: DoubleCCA: Improving Foundation Model Group Robustness with Random Sentence Embeddings},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/KP3GTUON}},
  note         = {Machine review of arXiv:2411.16236}
}
read the original abstract

This paper presents a novel method to improve the robustness of foundation models to group-based biases. We propose a simple yet effective method, called DoubleCCA, that leverages random sentences and Canonical Correlation Analysis (CCA) to enrich the text embeddings of the foundation model. First, we generate various random sentences that augment the original prompts, which extends the original prompts with random words or character sequences. Second, we use an additional sentence embedding model to generate different text embeddings with respect to these random sentences. We then use CCA double twice to align the representations and reconstruct them back to the original representation space. We demonstrate the effectiveness of our method on a variety of tasks and datasets, showing that it outperforms existing methods in terms of both performance and robustness. Our method is simple to implement and can be easily integrated into existing models, making it a practical solution for improving the robustness of foundation models to group-based biases.

Figures

Figures reproduced from arXiv: 2411.16236 by the authors.

Figure 1
Figure 1. The pipeline of our proposed DoubleCCA. We leverage random words to augment semantic descriptions and introduce an additional sentence embedding model to complement the semantic limitations of the original VLM text encoder. We use classical CCA technique double twice to merge different semantic information, which helps to improve the group robustness of the CLIP model. dataset, the background of the image is a group… view at source ↗
Figure 2
Figure 2. We compare the performance of different prompts with different backbone models on the Waterbirds dataset. “Ori” denotes the original prompt of CLIP, i.e., “a photo of a ⟨class name⟩”. “Waffle-1” denotes the combination of the original prompt and the random words, i.e., “a photo of a ⟨class name⟩, which has ⟨random word⟩”. “Waffle-2” also denotes the combination of the original prompt and the random words, but with d… view at source ↗
Figure 3
Figure 3. The visualization of the image embeddings of the Waterbirds dataset. We also visualize the text embedding features extracted by the CLIP text encoder. The “Ori prompt” means the original prompt, i.e., “a photo of a ⟨class name⟩”. The “Waffle prompt” denote the prompts with the random words and characters. 3.2 DoubleCCA According to the analysis in Sect.3.1, we argue that random words or characters will introduce ran… view at source ↗
Figures from the paper (3 more)
Figure 4
Figure 4. Figure 4: Combination of Contrastive Adapter (CA) and our proposed DoubleCCA. We report the average accuracy and worst group robustness on the Waterbirds dataset. The backbone model is ViT-L/14 and ResNet-50. on the CelebA dataset. For instance, when the backbone is ViT-B/16, th…
Figure 5
Figure 5. Figure 5: Ablation study results on the Waterbirds dataset. 1 10 50 100 200 500 1000 2000 Number 20 30 40 50 60 70 80 90 Percentage (%) Accuracy Worst 1 10 16 32 64 128 256 384 Dimension 20 30 40 50 60 70 80 90 Percentage (%) Accuracy Worst Avg Worst 20 40 60 80 100 Accuracy (%)…
Figure 6
Figure 6. Figure 6: Ablation study results on the Waterbirds dataset. BART shows little improvement in model performance. We think this is because BART targets dialogue understanding, question answering, and summarization tasks, which may face the same problems as mentioned before, where …

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

41 extracted references · 26 canonical work pages

  1. [1]

    Learning transferable visual models from natural language supervision

    Alec Radford, Jong Wook Kim, Chris Hallacy, Aditya Ramesh, Gabriel Goh, Sandhini Agarwal, Girish Sastry, Amanda Askell, Pamela Mishkin, Jack Clark, et al. Learning transferable visual models from natural language supervision. In ICML, 2021

    2021

  2. [2]

    Sigmoid loss for language image pre-training

    Xiaohua Zhai, Basil Mustafa, Alexander Kolesnikov, and Lucas Beyer. Sigmoid loss for language image pre-training. In ICCV, 2023

    2023

  3. [3]

    Hyperbolic image-text representations

    Karan Desai, Maximilian Nickel, Tanmay Rajpurohit, Justin Johnson, and Shanmukha Ramakrishna Vedantam. Hyperbolic image-text representations. In ICML, 2023

    work page 2023

  4. [4]

    Laion-400m: Open dataset of clip-filtered 400 million image-text pairs

    Christoph Schuhmann, Richard Vencu, Romain Beaumont, Robert Kaczmarczyk, Clayton Mullis, Aarush Katta, Theo Coombes, Jenia Jitsev, and Aran Komatsuzaki. Laion-400m: Open dataset of clip-filtered 400 million image-text pairs. arXiv:2111.02114, 2021

    arXiv 2021

  5. [5]

    Medclip: Contrastive learning from unpaired medical images and text

    Zifeng Wang, Zhenbang Wu, Dinesh Agarwal, and Jimeng Sun. Medclip: Contrastive learning from unpaired medical images and text. arXiv:2210.10163, 2022

    arXiv 2022

  6. [6]

    Geode: a geographically diverse evaluation dataset for object recognition

    Vikram V Ramaswamy, Sing Yu Lin, Dora Zhao, Aaron Adcock, Laurens van der Maaten, Deepti Ghadiyaram, and Olga Russakovsky. Geode: a geographically diverse evaluation dataset for object recognition. NeurIPS, 2024. 15

    work page 2024

  7. [7]

    Clip as rnn: Segment countless visual concepts without training endeavor

    Shuyang Sun, Runjia Li, Philip Torr, Xiuye Gu, and Siyang Li. Clip as rnn: Segment countless visual concepts without training endeavor. In CVPR, 2024

    work page 2024

  8. [8]

    Cascade-clip: Cascaded vision-language embeddings alignment for zero-shot semantic segmentation

    Yunheng Li, Zhong-Yu Li, Quan-Sheng Zeng, Qibin Hou, and Ming-Ming Cheng. Cascade-clip: Cascaded vision-language embeddings alignment for zero-shot semantic segmentation. In ICML, 2024

    work page 2024

Show all 41 references
  1. [9]

    Visual classification via description from large language models

    Sachit Menon and Carl Vondrick. Visual classification via description from large language models. In ICLR, 2022

    2022

  2. [10]

    Waffling around for performance: Visual classification with random words and broad concepts

    Karsten Roth, Jae Myung Kim, A Koepke, Oriol Vinyals, Cordelia Schmid, and Zeynep Akata. Waffling around for performance: Visual classification with random words and broad concepts. In ICCV, 2023

    2023

  3. [11]

    PerceptionCLIP: Visual classification by inferring and conditioning on contexts

    Bang An, Sicheng Zhu, Michael-Andrei Panaitescu-Liess, Chaithanya Kumar Mummadi, and Furong Huang. PerceptionCLIP: Visual classification by inferring and conditioning on contexts. In ICLR, 2024

    2024

  4. [12]

    Contrastive adapters for foundation model group robustness

    Michael Zhang and Christopher R ´e. Contrastive adapters for foundation model group robustness. NeurIPS, 2022

    2022

  5. [13]

    Fine-tuning can distort pretrained features and underperform out-of-distribution

    Ananya Kumar, Aditi Raghunathan, Robbie Matthew Jones, Tengyu Ma, and Percy Liang. Fine-tuning can distort pretrained features and underperform out-of-distribution. In ICLR, 2022

    2022

  6. [14]

    Last layer re-training is sufficient for robustness to spurious correlations

    Polina Kirichenko, Pavel Izmailov, and Andrew Gordon Wilson. Last layer re-training is sufficient for robustness to spurious correlations. In ICLR, 2023

    2023

  7. [15]

    Debiasing vision-language models via biased prompts

    Ching-Yao Chuang, Varun Jampani, Yuanzhen Li, Antonio Torralba, and Stefanie Jegelka. Debiasing vision-language models via biased prompts. arXiv:2302.00070, 2023

    2023 arXiv

  8. [16]

    FairerCLIP: Debiasing CLIP’s zero-shot predictions using functions in RKHSs

    Sepehr Dehdashtian, Lan Wang, and Vishnu Boddeti. FairerCLIP: Debiasing CLIP’s zero-shot predictions using functions in RKHSs. In ICLR, 2024

    2024

  9. [17]

    Calibrating multi-modal representations: A pursuit of group robustness without annotations

    Chenyu You, Yifei Mint, Weicheng Dai, Jasjeet S Sekhon, Lawrence Staib, and James S Duncan. Calibrating multi-modal representations: A pursuit of group robustness without annotations. In CVPR, 2024

    2024

  10. [18]

    Clip-adapter: Better vision-language models with feature adapters

    Peng Gao, Shijie Geng, Renrui Zhang, Teli Ma, Rongyao Fang, Yongfeng Zhang, Hongsheng Li, and Yu Qiao. Clip-adapter: Better vision-language models with feature adapters. IJCV, 2024

    2024

  11. [19]

    Controllable prompt tuning for balancing group distributional robustness

    Hoang Phan, Andrew Gordon Wilson, and Qi Lei. Controllable prompt tuning for balancing group distributional robustness. In ICML, 2024

    2024

  12. [20]

    Debiasing vison-language models with text-only training

    Yunfan Yang, Chaoquan Jiang, Zhiyu Lin, Jinlin Xiao, Jiaming Zhang, and Jitao Sang. Debiasing vison-language models with text-only training. arXiv:2410.09365, 2024. 16

    2024 arXiv

  13. [21]

    Language models as hierarchy encoders

    Yuan He, Zhangdie Yuan, Jiaoyan Chen, and Ian Horrocks. Language models as hierarchy encoders. In NeurIPS, 2024

    2024

  14. [22]

    Eigenproblems in pattern recognition

    Eduardo Bayro Corrochano, Tijl De Bie, Nello Cristianini, and Roman Rosipal. Eigenproblems in pattern recognition. Handbook of Geometric Computing: Applications in Pattern Recognition, Computer Vision, Neuralcomputing, and Robotics, 2005

    2005

  15. [23]

    Harmony in diversity: Merging neural networks with canonical correlation analysis

    Stefan Horoi, Albert Manuel Orozco Camacho, Eugene Belilovsky, and Guy Wolf. Harmony in diversity: Merging neural networks with canonical correlation analysis. In ICML, 2024

    2024

  16. [24]

    Hashimoto, and Percy Liang

    Shiori Sagawa, Pang Wei Koh, Tatsunori B. Hashimoto, and Percy Liang. Distributionally robust neural networks. In ICLR, 2020

    2020

  17. [25]

    Visualizing data using t-sne

    Laurens van der Maaten and Geoffrey Hinton. Visualizing data using t-sne. JMLR, 2008

    2008

  18. [26]

    Making monolingual sentence embeddings multilingual using knowledge distillation

    Nils Reimers and Iryna Gurevych. Making monolingual sentence embeddings multilingual using knowledge distillation. In EMNLP, 2020

    2020

  19. [27]

    Deep learning face attributes in the wild

    Ziwei Liu, Ping Luo, Xiaogang Wang, and Xiaoou Tang. Deep learning face attributes in the wild. In ICCV, 2015

    2015

  20. [28]

    The caltech-ucsd birds-200-2011 dataset

    Catherine Wah, Steve Branson, Peter Welinder, Pietro Perona, and Serge Belongie. The caltech-ucsd birds-200-2011 dataset. 2011

    2011

  21. [29]

    Eurosat: A novel dataset and deep learning benchmark for land use and land cover classification

    Patrick Helber, Benjamin Bischke, Andreas Dengel, and Damian Borth. Eurosat: A novel dataset and deep learning benchmark for land use and land cover classification. IEEE J. Sel. Top. Appl. Earth Obs. Remote Sens, 2019

    2019

  22. [30]

    Places: A 10 million image database for scene recognition

    Bolei Zhou, Agata Lapedriza, Aditya Khosla, Aude Oliva, and Antonio Torralba. Places: A 10 million image database for scene recognition. IEEE TPAMI, 2017

    2017

  23. [31]

    Automated flower classification over a large number of classes

    Maria-Elena Nilsback and Andrew Zisserman. Automated flower classification over a large number of classes. In ICVGIP, 2008

    2008

  24. [32]

    Food-101–mining discriminative components with random forests

    Lukas Bossard, Matthieu Guillaumin, and Luc Van Gool. Food-101–mining discriminative components with random forests. In ECCV, 2014

    2014

  25. [33]

    Cats and dogs

    Omkar M Parkhi, Andrea Vedaldi, Andrew Zisserman, and CV Jawahar. Cats and dogs. In CVPR, 2012

    2012

  26. [34]

    Pytorch: An imperative style, high- performance deep learning library

    Adam Paszke, Sam Gross, Francisco Massa, Adam Lerer, James Bradbury, Gregory Chanan, Trevor Killeen, Zeming Lin, Natalia Gimelshein, Luca Antiga, et al. Pytorch: An imperative style, high- performance deep learning library. NeurIPS, 2019. 17

    2019

  27. [35]

    Deep residual learning for image recognition

    Kaiming He, Xiangyu Zhang, Shaoqing Ren, and Jian Sun. Deep residual learning for image recognition. In CVPR, 2016

    2016

  28. [36]

    An image is worth 16x16 words: Transformers for image recognition at scale

    Alexey Dosovitskiy, Lucas Beyer, Alexander Kolesnikov, Dirk Weissenborn, Xiaohua Zhai, Thomas Unterthiner, Mostafa Dehghani, Matthias Minderer, Georg Heigold, Sylvain Gelly, Jakob Uszkoreit, and Neil Houlsby. An image is worth 16x16 words: Transformers for image recognition at...

    2021

  29. [37]

    Hyperbolic representation learning: Revisiting and advancing

    Menglin Yang, Min Zhou, Rex Ying, Yankai Chen, and Irwin King. Hyperbolic representation learning: Revisiting and advancing. In ICML, 2023

    2023

  30. [38]

    Utility-fairness trade-offs and how to find them

    Sepehr Dehdashtian, Bashir Sadeghi, and Vishnu Naresh Boddeti. Utility-fairness trade-offs and how to find them. In CVPR, 2024

    2024

  31. [39]

    Towards general text embeddings with multi-stage contrastive learning

    Zehan Li, Xin Zhang, Yanzhao Zhang, Dingkun Long, Pengjun Xie, and Meishan Zhang. Towards general text embeddings with multi-stage contrastive learning. arXiv:2308.03281, 2023

    2023 arXiv

  32. [40]

    BART: Denoising sequence-to-sequence pre-training for natural language generation, translation, and comprehension

    Mike Lewis, Yinhan Liu, Naman Goyal, Marjan Ghazvininejad, Abdelrahman Mohamed, Omer Levy, Veselin Stoyanov, and Luke Zettlemoyer. BART: Denoising sequence-to-sequence pre-training for natural language generation, translation, and comprehension. In ACL, 2020

    2020

  33. [41]

    Principal component analysis

    Ren´e Vidal, Yi Ma, S Shankar Sastry, Ren´e Vidal, Yi Ma, and S Shankar Sastry. Principal component analysis. Generalized principal component analysis, 2016. 18

    2016

Pith tools

Reviewed August 12, 2026 · model on record in the stance chip above.