Pith. sign in

REVIEW 4 major objections 6 minor 54 references

Enhancing VICReg: Random-Walk Pairing for Improved Generalization and Better Global Semantics Capturing

T0 review · 4 major / 6 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 Pairing each image with a random-walk neighbor in embedding space lets VICReg preserve global semantic structure on classes it never saw during training.

desk verdict Plausible drop-in VICReg enhancement with a new label-free metric, but the headline semantic-generalization claim rests on an unvalidated metric that shares geometry with the training loss. read the letter →

arxiv 2506.18104 v1 pith:Z65NXRAJ submitted 2025-06-22 cs.CV cs.LG

Idan Simai , Ronen Talmon , Uri Shaham This is my paper · ORCID
classification cs.CVcs.LG
keywords self-supervisedlearningVICRegspectralembeddingrandom-walkpairingglobalsemanticshierarchicalrepresentationlabel-freeevaluationgeneralization
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 argues that the popular self-supervised method VICReg behaves like a spectral embedding on a graph whose only edges link augmentations of the same image. Because spectral embeddings are built for a fixed graph, a model trained this way can behave erratically when asked to embed images from classes it never saw during training. To fix this, the paper introduces SAG-VICReg, which adds pairs of different but similar images to the invariance loss by sampling a random walk on a nearest-neighbor affinity matrix and weighting each pair by embedding similarity. The paper also proposes a label-free evaluation metric that compares the hierarchical trees (dendrograms) of two embedding sets, and reports that SAG-VICReg matches or outperforms leading self-supervised baselines on this metric across ImageNet, CIFAR-100, and Caltech-256 while keeping competitive classification accuracy.

What carries the argument

The load-bearing object is the random-walk pairing over a batch-level similarity graph. For each anchor embedding z_i in one augmented batch, the paper builds a sparse affinity matrix W_ij = exp(-$d_ij^{2}$ / $sigma_i^{2}$) for the k = 5 nearest neighbors of z_i in the other batch, with sigma_i set by the 20th percentile of adjusted distances, then samples a partner z''_i according to the row-stochastic random-walk matrix P = $D^{-1}$ W. The weighted invariance term, with variance and covariance terms unchanged, is what densifies the augmentation graph and, according to the paper's theoretical appendix, makes the embedding smoother on unseen regions.

What would settle it

Run the paper's LCA and cophenetic protocol on a deliberately non-semantic control: train two models on disjoint class splits with the SAG-VICReg pipeline but replace the similarity-based affinity matrix with a random permutation of the k-nearest-neighbor graph. If the control's dendrogram correlations stay close to SAG-VICReg's reported values, the metric is not measuring semantics; if the control drops sharply, also test several random class splits to confirm the reported gains are not an artifact of always evaluating the first half of the class list.

Watch

Extended reading notes

Core claim

On its own terms, the paper claims that VICReg's instability on unseen classes follows from its spectral-embedding interpretation: the invariance term minimizes a weighted sum of squared embedding distances with weights that are nonzero only for augmentations of the same image, so the learned map is not constrained outside the training clusters. SAG-VICReg replaces that all-or-nothing weight matrix with a Gaussian affinity built from the k nearest neighbors by cosine similarity, constructs the random-walk transition matrix P = $D^{-1}$ W, samples one partner z''_i from the second augmented batch per anchor z_i, and weights the invariance loss by the corresponding similarity W_ij. The paper reports that this densification of cross-image edges yields higher LCA and cophenetic correlations between embeddings of the same images produced by models trained on disjoint class splits, and higher hierarchical Rand indices, while retaining linear and k-NN accuracy at fine grain.

Load-bearing premise

The paper's evidence that SAG-VICReg captures global semantics rests on the assumption that the new label-free LCA and cophenetic correlations between dendrograms of two models trained on disjoint classes measure genuine semantic hierarchy, rather than generic structural similarity that any reasonable embedding would show.

Editorial extensions

If this is right

  • If SAG-VICReg's reported LCA and cophenetic gains hold, self-supervised encoders can be made stable for entire unseen classes, not just unseen instances of seen classes, without changing the backbone or adding memory banks.
  • Linear and k-NN probes can rank models differently from hierarchy-aware metrics, so using both together gives a more complete picture of representation quality when labels are scarce.
  • The random-walk pairing strategy can be dropped into other invariance-based SSL losses by changing only which pairs enter the invariance term.
  • Higher hierarchical Rand indices at coarse levels imply the method helps when the downstream task needs grouping, such as clustering or semi-supervised classification with few labels.
  • The label-free metric can be reused to monitor whether a deployed embedding model keeps its structure under distribution shift, since it only needs two embedding sets of the same entities.

Reading between the lines

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

  • An extension the paper leaves implicit is that the same random-walk pairing could be applied with a memory bank or queue, so pairs are drawn from a much larger and more global neighborhood, which may further strengthen the reported out-of-cluster stability.
  • A natural next step is to use the LCA and cophenetic correlations not only for evaluation but as a model-selection criterion for choosing augmentations or training hyperparameters, since the metric needs no labels.
  • If the metric genuinely tracks semantic hierarchy, it should predict performance on unseen-class transfer: a model with high train/test dendrogram correlation should also do better in a zero-shot-style retrieval or coarse-level clustering probe, a directly testable consequence the paper does not report.
  • The paper's split protocol always tests on the first half of the class list while training on first versus second halves; checking multiple random class splits would show whether the reported gains depend on that particular class ordering.
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 / 6 minor

Summary. The paper argues that VICReg can be viewed as an instance of SpectralNet on a graph whose clusters are sets of augmentations of the same image, and that this spectral viewpoint exposes a generalization weakness when embeddings are computed for images from unseen classes. To address this, the authors propose SAG-VICReg, which augments the VICReg invariance loss by sampling cross-image pairs from a random walk on a cosine-similarity affinity matrix and weighting those pairs by their similarity. The paper also introduces a label-free evaluation metric based on the correlation of hierarchical dendrograms (LCA distances and cophenetic distances) between two embedding sets, and uses this metric, together with hierarchical Rand indices and hierarchical linear/k-NN classification, to claim that SAG-VICReg better captures global semantic structure while maintaining local discriminative performance. Experiments are reported on ImageNet-1k, CIFAR-100, and Caltech-256, with comparisons against VICReg, SimCLR, Barlow Twins, DINO, MAE, and I-JEPA.

Significance. If the central claims hold, the paper makes two useful contributions: a training modification that improves out-of-class generalization of invariance-based SSL, and a label-free evaluation protocol for global structure that could complement linear-probe and k-NN evaluation. The paper is explicit about its algorithm, hyperparameters, and data splits, and it evaluates on multiple datasets and multiple hierarchy-aware metrics, which is a strength. The theoretical motivation through spectral embedding is interesting and connects to an existing literature on augmentation graphs. However, the headline evidence for 'global semantic understanding' rests on a new metric that is not calibrated against semantic ground truth or null models, and the training objective itself depends on the current embedding geometry, so the possibility that the measured gains reflect self-consistency rather than true semantic generalization is a genuine, unresolved concern.

major comments (4)
  1. [Section 4.4 and Appendix E, Tables 1–3] The label-free LCA/cophenetic metric is not validated against semantic ground truth or any null model. In the protocol of Appendix E, the metric compares dendrograms of embeddings of the same test images produced by two models trained on disjoint class sets. A high correlation between the two dendrograms shows that the two models organize the test images similarly, but it does not by itself show that either organization matches true semantic hierarchy. Because the training objective (Section 4.3, Algorithm 1) builds its affinity graph from cosine nearest neighbors and the evaluation metric also builds cosine-distance dendrograms, a model that simply produces internally consistent cosine neighborhoods could inflate Tables 1–3 without improving semantics. The reported correlations are only 0.2–0.45, and no control is provided using a randomly initialized network, a label-shuffled model, or an embedding space with a deliberately wrong semantic structure. I ask the authors to add a label-anchored validation (e.g., correlate the learned dendrograms with the known class hierarchy at multiple levels) and null-model baselines, and to show that the metric separates semantically correct from semantically arbitrary embeddings.
  2. [Section 4.3, Algorithm 1, and Section 5] The pair-sampling mechanism is self-referential. The affinity matrix W is constructed from the model's own current embeddings z and z', and the random walk then samples positive pairs according to that affinity. The model is therefore trained to make its embeddings invariant to the similarity structure that it already produces. This is a common bootstrap in SSL, but it creates a risk that the improvements in Tables 1–3 reflect the model becoming more consistent with its own nearest-neighbor geometry rather than learning external semantic structure. The paper would be substantially strengthened by an ablation that fixes the graph W (e.g., from a pretrained feature extractor or from a random subset) or that compares random-walk sampling against random cross-image pairs with the same weighting. Without such a control, the causal claim that random-walk pairing 'captures global semantics' is not fully established.
  3. [Section 4.1 and Appendix G.1] The claimed equivalence between VICReg and SpectralNet is approximate and should be stated with its limitations. VICReg's variance and covariance regularizers are soft penalties, whereas SpectralNet enforces an exact orthogonality constraint through an orthogonalization layer; the invariance loss in VICReg uses W with entries in {0,1} and is computed on pairs of augmentations, not on all pairs of distinct data points. The paper uses this equivalence to motivate the method, and the text in Section 4.1 and the caption of Figure 1 suggest a stronger formal connection than is actually proved. I would like the authors to state precisely under which conditions the VICReg loss approximates the Laplacian eigenproblem, and to acknowledge that the theoretical justification in Appendix B cites related results but does not prove the claimed generalization improvement for SAG-VICReg.
  4. [Tables 2 and 3 and surrounding text] The text overstates the consistency of the empirical gains. In Table 2, the D2-to-P1 column on ImageNet reports a negative relative gain (-5.28%) for SAG-VICReg against VICReg, yet the text says SAG-VICReg 'again demonstrates superior performance.' In Table 3, SAG-VICReg has the highest Pearson on CIFAR-100 only for I-JEPA (0.325), not for SAG-VICReg (0.201); the sentence 'our enhancements elevate VICReg from the worst-performing method ... to the best-performing approach in capturing global semantic structure' is not supported by the Pearson column, where SAG-VICReg is below I-JEPA. These discrepancies should be acknowledged, and the claims should be limited to the specific metrics and comparisons where the gains actually hold.
minor comments (6)
  1. [General] The manuscript repeatedly uses 'VICREG' (e.g., Appendix J) instead of 'VICReg'; please standardize the spelling.
  2. [Section 4.1] The phrase 'as it is excepted to' should read 'as it is expected to'.
  3. [Section 4.2] There is a typo in 'global strcture' and an ungrammatical phrase 'as can be seen in The bottom row of Figure 5'; please edit.
  4. [Appendix G.3] The sentence 'The distance metric of The distance metric of D can be Euclidean or cosine distance' contains a duplicated phrase; please correct.
  5. [Figure 4] Figure 4 reports hierarchical Rand indices without error bars or an explicit statement of the number of runs; adding variability estimates would make the claimed robustness more convincing.
  6. [Appendix E] The paper does not state how many independent training runs or random seeds underlie the reported standard deviations in Tables 1–5; please report the number of repetitions per experiment.

Circularity Check

1 steps flagged · score 4.0 of 10

The new label-free LCA/cophenetic metric shares the cosine-similarity geometry that the SAG-VICReg training loss explicitly optimizes, so the headline global-semantics gains in Tables 1–3 are partly self-consistency by construction; label-based results provide some independent support.

  1. self definitional [Section 4.3 (Algorithm 1, weighted invariance loss) and Section 4.4 (LCA/cophenetic evaluation); Tables 1–3]
    "Compute affinity matrix W∈R^{n×n} between samples in Z and Z′ ... Compute a Random Walk matrix P and sample n pairs to form the new batch Z′′ ... Compute the weighted VICReg’s invariance loss using W_ij between Z,Z′′ (3)"

    The affinity matrix W is built from the model's own cosine-similarity nearest-neighbor graph, and the random-walk pairs are sampled from that same graph. The invariance loss then pulls together exactly those pairs. The evaluation metric in Section 4.4 constructs dendrograms using cosine distance and Ward linkage, and reports LCA/cophenetic correlations between such dendrograms as evidence of 'global semantic structure' and 'generalization ability.' Thus the training objective directly shapes the cosine-hierarchy consistency that Tables 1–3 score; the label-free metric is never calibrated against known hierarchies or a null model, so the reported gains partly measure self-consistency with the training objective rather than independently established semantic correctness.

full rationale

The paper's reformulation of VICReg as a spectral embedding method is a legitimate mathematical reinterpretation, not a circular derivation: the invariance term with a diagonal W and the variance/covariance terms are rearranged into a SpectralNet-like loss, and the generalization concern is illustrated with independent toy and real-data examples. The random-walk pairing is a genuine algorithmic modification, and the hyperparameters k and scale are disclosed rather than silently fitted to the headline metric. The main circularity concern is evaluative: the proposed label-free LCA/cophenetic metric and the SAG-VICReg training loss both operate on cosine-similarity neighborhoods of the model's own embeddings. Consequently, the 'global semantic understanding' improvements in Tables 1–3 partly reflect the model becoming more self-consistent with the geometry it was trained to produce, rather than being validated against an external semantic hierarchy. No load-bearing self-citation was found: the SpectralNet citation is prior external work, and Appendix B cites independent theoretical results. The label-based Rand index and hierarchical linear/k-NN classification provide independent evidence that the method retains discriminative quality, so the circularity is partial rather than total.

Assumptions & free parameters 2 free parameters · 5 assumptions · 0 invented entities

The central claim rests on two tuned hyperparameters, k and scale, and on five domain assumptions. No invented entities are introduced. The main burden is that both the training signal (random-walk pairs) and the headline evaluation signal (LCA/cophenetic correlations) are derived from the model's own embedding geometry or cross-model dendrogram agreement, with only the hierarchical Rand index and classification tables providing external semantic grounding.

free parameters (2)
  • k (number of nearest neighbors) = 5
    Chosen as consistently optimal across datasets and metrics in Appendix C; no separate held-out tuning is reported.
  • Gaussian kernel scale (per-point) = 20th percentile of adjusted distances, clamped at 1e-7
    Tuned for best balance of performance and stability in Appendix C and Appendix I; it is a data-dependent hyperparameter.
assumptions (5)
  • domain assumption Soft variance and covariance regularizers in VICReg enforce the same orthonormal output constraint as SpectralNet's exact orthogonalization layer.
    Used in Section 4.1 with footnote 1 to identify VICReg as spectral embedding; the regularizers are soft penalties, so the equivalence is approximate.
  • domain assumption Failure of classical spectral embeddings to generalize to new clusters transfers to deep VICReg embeddings of unseen classes.
    The motivation of Figures 1 and 2; the 2D example and qualitative image similarity do not quantitatively establish this transfer.
  • domain assumption Cosine-similarity neighborhoods in the current embedding space correspond to true semantic relationships.
    Section 4.3 builds the affinity and random-walk graph from the model's own embeddings; the method assumes these estimated edges are semantically correct and that reinforcing them improves generalization.
  • domain assumption Dendrogram correlation between embeddings from two models trained on disjoint class splits measures global semantic generalization.
    Section 4.4 and Appendix E; this is the load-bearing evaluation premise behind Tables 1-3 and is not validated against semantic ground truth.
  • domain assumption Theoretical denser-edge guarantees from contrastive learning apply to SAG-VICReg's self-estimated random-walk edges.
    Appendix B invokes HaoChen et al., Li et al., and Cabannes et al.; the cited theory concerns augmentation graphs, not edges estimated from a model's own embeddings.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Enhancing VICReg: Random-Walk Pairing for Improved Generalization and Better Global Semantics Capturing." pith.science (2026). https://pith.science/paper/Z65NXRAJ

@misc{pith2026250618104,
  author       = {Pith},
  title        = {Pith review of: Enhancing VICReg: Random-Walk Pairing for Improved Generalization and Better Global Semantics Capturing},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/Z65NXRAJ}},
  note         = {Machine review of arXiv:2506.18104}
}
read the original abstract

In this paper, we argue that viewing VICReg-a popular self-supervised learning (SSL) method--through the lens of spectral embedding reveals a potential source of sub-optimality: it may struggle to generalize robustly to unseen data due to overreliance on the training data. This observation invites a closer look at how well this method achieves its goal of producing meaningful representations of images outside of the training set as well. Here, we investigate this issue and introduce SAG-VICReg (Stable and Generalizable VICReg), a method that builds on VICReg by incorporating new training techniques. These enhancements improve the model's ability to capture global semantics within the data and strengthen the generalization capabilities. Experiments demonstrate that SAG-VICReg effectively addresses the generalization challenge while matching or surpassing diverse state-of-the-art SSL baselines. Notably, our method exhibits superior performance on metrics designed to evaluate global semantic understanding, while simultaneously maintaining competitive results on local evaluation metrics. Furthermore, we propose a new standalone evaluation metric for embeddings that complements the standard evaluation methods and accounts for the global data structure without requiring labels--a key issue when tagged data is scarce or not available.

Figures

Figures reproduced from arXiv: 2506.18104 by the authors.

Figure 1
Figure 1. A real demonstration of the unpredictable behavior of spectral embeddings when encountering new points from an unseen cluster during the test. In particular, Subfigure (a) shows the original training data with 3 clusters (green, purple, orange). Subfigure (b) displays the test data with the same 3 clusters plus a new blue cluster. Subfigure (c) shows how each of the original training clusters is embedded to a point.… view at source ↗
Figure 2
Figure 2. A real-world demonstration of how VICReg behaves unstably when encountering images from unseen clusters during test time. We compare cosine similarity scores between the query image (leftmost) and other images, showing results from both VICReg and our approach. Notice how VICReg’s embeddings can place semantically similar images far apart and semantically unrelated images too close together—an empirical demonstratio… view at source ↗
Figure 3
Figure 3. Non-trivial pairs. For each column, the upper image shows where the random walk starts and the lower image shows where it ends. This sampling process creates a new batch of pairs (zi , z′′ i ), where z ′′ i is selected from Z ′ using the the Random Walk P starting from zi . Finally, we compute the invariance term in the VICReg loss, weighted by the similarity values Wij (3). Specifically, the invariance constraint i… view at source ↗
Figures from the paper (3 more)
Figure 4
Figure 4. Figure 4: Comparison of Rand-Index scores across hierarchy levels (left) and varying numbers of clusters (right). [PITH_FULL_IMAGE:figures/full_fig_p008_4.png]
Figure 5
Figure 5. Figure 5: Comparing embedding quality assessment methods. Top: Original data (left) and two embeddings, where the first (middle) is linearly separable but distorts global structure, while the second (right) preserves structural relationships but isn’t linearly separable. Bottom:…
Figure 6
Figure 6. Figure 6: UMAP visualizations of the class centroids (mean embeddings) for SAG-VICReg and VICReg. Both models were trained on the same set of classes, which differ from those used to generate the embedded images. SAG-VICReg more effectively preserves global structure even on uns…

Review history (2 revisions) →

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

54 extracted references · 37 canonical work pages

  1. [14]

    Contrastive and non-contrastive self-supervised learning recover global and local spectral embedding methods

    Randall Balestriero and Yann LeCun. Contrastive and non-contrastive self-supervised learning recover global and local spectral embedding methods. InAdvances in Neural Information Processing Systems, volume 35, pages 26671–26685. Curran Associates, Inc., 2022

    work page 2022

  2. [1]

    Masked autoencoders are scalable vision learners.Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 16000–16009, 2022

    Kaiming He, Xinlei Chen, Saining Xie, Yanghao Li, Piotr Dollár, and Ross Girshick. Masked autoencoders are scalable vision learners.Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 16000–16009, 2022

    work page 2022

  3. [2]

    Masked siamese networks for label-efficient learning.European Conference on Computer Vision, pages 431–447, 2022

    Mahmoud Assran, Mathilde Caron, Ishan Misra, Piotr Bojanowski, Armand Joulin, Nicolas Ballas, and Michael Rabbat. Masked siamese networks for label-efficient learning.European Conference on Computer Vision, pages 431–447, 2022

    work page 2022

  4. [3]

    Self-supervised learning from images with a joint-embedding predictive architecture.arXiv preprint arXiv:2301.08243, 2023

    Mahmoud Assran, Quentin Duval, Ishan Misra, Piotr Bojanowski, Pascal Vincent, Michael Rabbat, Yann LeCun, and Nicolas Ballas. Self-supervised learning from images with a joint-embedding predictive architecture.arXiv preprint arXiv:2301.08243, 2023

    arXiv 2023

  5. [4]

    Learning and leveraging world models in visual representation learning.arXiv preprint arXiv:2403.00504, 2024

    Quentin Garrido, Mahmoud Assran, Nicolas Ballas, Adrien Bardes, Laurent Najman, and Yann LeCun. Learning and leveraging world models in visual representation learning.arXiv preprint arXiv:2403.00504, 2024

    arXiv 2024

  6. [5]

    Exploring simple siamese representation learning

    Xinlei Chen and Kaiming He. Exploring simple siamese representation learning. InProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), 2020

    work page 2020

  7. [6]

    Whitening for self-supervised representation learning.arXiv preprint arXiv:2007.06346, 2021

    Aleksandr Ermolov, Aliaksandr Siarohin, Enver Sangineto, and Nicu Sebe. Whitening for self-supervised representation learning.arXiv preprint arXiv:2007.06346, 2021

    arXiv 2007

  8. [7]

    Obow: Online bag-of-visual-words generation for self-supervised learning

    Spyros Gidaris, Andrei Bursuc, Gilles Puy, Nikos Komodakis, Matthieu Cord, and Patrick Pérez. Obow: Online bag-of-visual-words generation for self-supervised learning. InProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), 2021

    work page 2021

Show all 54 references
  1. [8]

    What makes for good views for contrastive learning

    Yonglong Tian, Chen Sun, Ben Poole, Dilip Krishnan, Cordelia Schmid, and Phillip Isola. What makes for good views for contrastive learning. InAdvances in Neural Information Processing Systems (NeurIPS), 2020

    2020

  2. [9]

    A simple framework for contrastive learning of visual representations

    Ting Chen, Simon Kornblith, Mohammad Norouzi, and Geoffrey Hinton. A simple framework for contrastive learning of visual representations. InInternational conference on machine learning, pages 1597–1607. PMLR, 2020

    2020

  3. [10]

    Barlow twins: Self-supervised learning via redundancy reduction

    Jure Zbontar, Li Jing, Ishan Misra, Yann LeCun, and Stéphane Deny. Barlow twins: Self-supervised learning via redundancy reduction. InInternational Conference on Machine Learning, pages 12310–12320. PMLR, 2021. 9

    2021

  4. [11]

    Emerging properties in self-supervised vision transformers

    Mathilde Caron, Hugo Touvron, Ishan Misra, Hervé Jégou, Julien Mairal, Piotr Bojanowski, and Armand Joulin. Emerging properties in self-supervised vision transformers. InProceedings of the IEEE/CVF International Conference on Computer Vision (ICCV), pages 9650–9660, 2021

    2021

  5. [12]

    Vicreg: Variance-invariance-covariance regularization for self- supervised learning

    Adrien Bardes, Jean Ponce, and Yann LeCun. Vicreg: Variance-invariance-covariance regularization for self- supervised learning. InInternational Conference on Learning Representations, 2021

    2021

  6. [13]

    Laplacian eigenmaps for dimensionality reduction and data representation

    Mikhail Belkin and Partha Niyogi. Laplacian eigenmaps for dimensionality reduction and data representation. Neural computation, 15(6):1373–1396, 2003

    2003

  7. [15]

    A tutorial on spectral clustering.Statistics and computing, 17(4):395–416, 2007

    Ulrike von Luxburg. A tutorial on spectral clustering.Statistics and computing, 17(4):395–416, 2007

    2007

  8. [16]

    Diffusion maps.Applied and Computational Harmonic Analysis, 21(1):5–30, 2006

    Ronald R Coifman and St’ephane Lafon. Diffusion maps.Applied and Computational Harmonic Analysis, 21(1):5–30, 2006

    2006

  9. [17]

    On spectral clustering: Analysis and an algorithm

    Andrew Y Ng, Michael I Jordan, and Yair Weiss. On spectral clustering: Analysis and an algorithm. InAdvances in neural information processing systems, pages 849–856, 2002

    2002

  10. [18]

    Spectralnet: Spectral clustering using deep neural networks

    Uri Shaham, Kelly Stanton, Henry Li, Boaz Nadler, Ronen Basri, and Yuval Kluger. Spectralnet: Spectral clustering using deep neural networks. In6th International Conference on Learning Representations, ICLR 2018, 2018

    2018

  11. [19]

    Learning multiple layers of features from tiny images

    Alex Krizhevsky and Geoffrey Hinton. Learning multiple layers of features from tiny images. 2009

    2009

  12. [20]

    Imagenet: A large-scale hierarchical image database.2009 IEEE Conference on Computer Vision and Pattern Recognition, pages 248–255, 2009

    Jia Deng, Wei Dong, Richard Socher, Li-Jia Li, Kai Li, and Li Fei-Fei. Imagenet: A large-scale hierarchical image database.2009 IEEE Conference on Computer Vision and Pattern Recognition, pages 248–255, 2009

    2009

  13. [21]

    Geometric harmonics: a novel tool for multiscale out-of-sample extension of empirical functions.Applied and Computational Harmonic Analysis, 21(1):31–52, 2006

    Ronald R Coifman and Stephane Lafon. Geometric harmonics: a novel tool for multiscale out-of-sample extension of empirical functions.Applied and Computational Harmonic Analysis, 21(1):31–52, 2006

    2006

  14. [22]

    Out-of-sample extensions for lle, isomap, mds, eigenmaps, and spectral clustering

    Yoshua Bengio, Jean-Francois Paiement, Pascal Vincent, Olivier Delalleau, Nicolas L Roux, and Marie Ouimet. Out-of-sample extensions for lle, isomap, mds, eigenmaps, and spectral clustering. InAdvances in neural information processing systems, pages 177–184, 2004

    2004

  15. [23]

    Spectral grouping using the nyström method

    Charless Fowlkes, Serge Belongie, Fan Chung, and Jitendra Malik. Spectral grouping using the nyström method. IEEE transactions on pattern analysis and machine intelligence, 26(2):214–225, 2004

    2004

  16. [24]

    Petros Drineas and Michael W. Mahoney. On the Nyström method for approximating a Gram matrix for improved kernel-based learning.Journal of Machine Learning Research, 6:2153–2175, 2005

    2005

  17. [25]

    Deep graph spectral evolution networks for graph topological evolution

    Negar Etemadyrad, Qingzhe Li, and Liang Zhao. Deep graph spectral evolution networks for graph topological evolution. InProceedings of the AAAI Conference on Artificial Intelligence, volume 35, pages 7827–7835, 2021

    2021

  18. [26]

    Contrastive multiview coding

    Yonglong Tian, Dilip Krishnan, and Phillip Isola. Contrastive multiview coding. InEuropean conference on computer vision, pages 776–794. Springer, 2020

    2020

  19. [27]

    Unsupervised learning of visual features by contrasting cluster assignments

    Mathilde Caron, Ishan Misra, Julien Mairal, Priya Goyal, Piotr Bojanowski, and Armand Joulin. Unsupervised learning of visual features by contrasting cluster assignments. InAdvances in Neural Information Processing Systems, pages 9912–9924, 2020

    2020

  20. [28]

    With a little help from my friends: Nearest-neighbor contrastive learning of visual representations

    Debidatta Dwibedi, Jonathan Tompson, Mohi Khansari, and Sergey Levine. With a little help from my friends: Nearest-neighbor contrastive learning of visual representations. InInternational Conference on Machine Learning, 2021

    2021

  21. [29]

    Hierarchical classification at multiple operating points.arXiv preprint arXiv:2210.10929, 2022

    Jack Valmadre et al. Hierarchical classification at multiple operating points.arXiv preprint arXiv:2210.10929, 2022

    2022 arXiv

  22. [30]

    Multi-level classification: A generic classification method for medical datasets

    M Srinivas, R Bharath, P Rajalakshmi, and C Krishna Mohan. Multi-level classification: A generic classification method for medical datasets. In2015 17th International Conference on E-health Networking, Application & Services (HealthCom), pages 262–267, 2015

    2015

  23. [31]

    Silla Jr and Alex A

    Carlos N. Silla Jr and Alex A. Freitas. A survey of hierarchical classification across different application domains. Data Mining and Knowledge Discovery, 22(1-2):31–72, 2010

    2010

  24. [32]

    Allan D. Gordon. Hierarchical classification. In P. Arabie, L. J. Hubert, and G. De Soete, editors,Clustering and Classification, pages 65–121. World Scientific Publishing, River Edge, NJ, 1996

    1996

  25. [33]

    Coherent hierarchical multi-label classification networks

    Eleonora Giunchiglia and Thomas Lukasiewicz. Coherent hierarchical multi-label classification networks. In Proceedings of the AAAI Conference on Artificial Intelligence, pages 2116–2123, New York, NY , 2020. AAAI Press. 10

    2020

  26. [34]

    Barros, and André C.P.L.F

    Ricardo Cerri, Rodrigo C. Barros, and André C.P.L.F. de Carvalho. Hierarchical multi-label classification using local neural networks.Journal of Computer and System Sciences, 80(1):39–56, 2014

    2014

  27. [35]

    Evaluation measures for hierarchical classification: A unified view and novel approaches

    Aris Kosmopoulos et al. Evaluation measures for hierarchical classification: A unified view and novel approaches. arXiv preprint arXiv:1511.07310, 2015

    2015 arXiv

  28. [36]

    Deep residual learning for image recognition

    Kaiming He, Xiangyu Zhang, Shaoqing Ren, and Jian Sun. Deep residual learning for image recognition. In Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition (CVPR), pages 770–778, 2016

    2016

  29. [37]

    An image is worth 16x16 words: Transformers for image recognition at scale.arXiv preprint arXiv:2010.11929, 2020

    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...

    2010 arXiv

  30. [38]

    Freeman, and Mark Hamilton

    Shaden Alshammari, John Hershey, Axel Feldmann, William T. Freeman, and Mark Hamilton. I-con: A unifying framework for representation learning. InProceedings of the International Conference on Learning Representations (ICLR), 2025. Also introduces neighbour–propagation debiasi...

    2025

  31. [39]

    L. L. McQuitty. Elementary linkage analysis for isolating orthogonal and oblique types and typal relevancies. Educational and Psychological Measurement, 17:207–209, 1957

    1957

  32. [40]

    M. R. Anderberg.Cluster Analysis for Applications. Academic Press, 1973

    1973

  33. [41]

    Caltech-256 object category dataset, March 2007

    Gregory Griffin, Alex Holub, and Pietro Perona. Caltech-256 object category dataset, March 2007. Available at http://www.vision.caltech.edu/Image_Datasets/Caltech256/

    2007

  34. [42]

    Hd-cnn: Hierarchical deep convolutional neural network for large scale visual recognition

    Zhicheng Yan, Hao Zhang, Robinson Piramuthu, Vignesh Jagadeesh, Dennis DeCoste, Wei Di, and Yizhou Yu. Hd-cnn: Hierarchical deep convolutional neural network for large scale visual recognition. InProceedings of the IEEE International Conference on Computer Vision (ICCV), pages...

    2015

  35. [43]

    YOLO9000: better, faster, stronger.CoRR, abs/1612.08242, 2016

    Joseph Redmon and Ali Farhadi. YOLO9000: better, faster, stronger.CoRR, abs/1612.08242, 2016

    2016 arXiv

  36. [44]

    Hyeongji Kim, Pekka Parviainen, Terje Berge, and Ketil Malde. Inspecting class hierarchies in classification-based metric learning models.Institute of Marine Research, Bergen, Norway, and Department of Informatics, University of Bergen, Norway, n.d

  37. [45]

    HaoChen, Colin Wei, Adrien Gaidon, and Tengyu Ma

    Jeff Z. HaoChen, Colin Wei, Adrien Gaidon, and Tengyu Ma. Provable guarantees for self-supervised deep learning with spectral contrastive loss. InAdvances in Neural Information Processing Systems, 2021

    2021

  38. [46]

    Understanding the generalization performance of spectral clustering algorithms

    Shaojie Li, Sheng Ouyang, and Yong Liu. Understanding the generalization performance of spectral clustering algorithms. InProceedings of the AAAI Conference on Artificial Intelligence (AAAI), 2023

    2023

  39. [47]

    Kiani, Randall Balestriero, Yann LeCun, and Alberto Bietti

    Vivien Cabannes, Bobak T. Kiani, Randall Balestriero, Yann LeCun, and Alberto Bietti. The ssl interplay: Augmentations, inductive bias, and generalization. InProceedings of the 40th International Conference on Machine Learning (ICML), 2023

    2023

  40. [48]

    Large batch training of convolutional networks.arXiv preprint arXiv:1708.03888, 2017

    Yang You, Igor Gitman, and Boris Ginsburg. Large batch training of convolutional networks.arXiv preprint arXiv:1708.03888, 2017

    2017 arXiv

  41. [49]

    Accurate, large minibatch sgd: Training imagenet in 1 hour.arXiv preprint arXiv:1706.02677, 2017

    Priya Goyal, Piotr Dollár, Ross Girshick, Pieter Noordhuis, Lukasz Wesolowski, Aapo Kyrola, Andrew Tulloch, Yangqing Jia, and Kaiming He. Accurate, large minibatch sgd: Training imagenet in 1 hour.arXiv preprint arXiv:1706.02677, 2017

    2017 arXiv

  42. [50]

    SGDR: Stochastic gradient descent with warm restarts

    Ilya Loshchilov and Frank Hutter. SGDR: Stochastic gradient descent with warm restarts. InInternational Conference on Learning Representations (ICLR), 2017

    2017

  43. [51]

    Decoupled weight decay regularization.arXiv preprint arXiv:1711.05101, 2017

    Ilya Loshchilov and Frank Hutter. Decoupled weight decay regularization.arXiv preprint arXiv:1711.05101, 2017

    2017 arXiv

  44. [52]

    Robert Sokal and F. Rohlf. Sokal rr, rohlf fj. the comparison of dendrograms by objective methods. taxon 11: 33-40.Taxon, 11:33–40, 02 1962

    1962

  45. [53]

    William M. Rand. Objective criteria for the evaluation of clustering methods.Journal of the American Statistical Association, 66(336):846–850, 1971

    1971

  46. [54]

    superman

    Leland McInnes, John Healy, and James Melville. Umap: Uniform manifold approximation and projection for dimension reduction.arXiv preprint arXiv:1802.03426, 2018. 11 A Visualizations Figure 5:Comparing embedding quality assessment methods.Top:Original data (left) and two embed...

    2018 arXiv

Pith tools

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