Pith. sign in

REVIEW 4 major objections 6 minor 51 references

Cluster Specific Representation Learning

T0 review · 4 major / 6 minor · reviewed 2026-08-11 · deepseek-v4-flash

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

Pith's one-line read A good representation is specific to each inherent cluster in the data, and a partially tensorized architecture makes this idea work across autoencoders, VAEs, contrastive learning, and RBMs.

desk verdict A small, clearly explained parameter-efficiency trick for tensorized representation learning, but the experiments don't isolate what causes the reported gains. read the letter →

arxiv 2412.03471 v1 pith:NPUPSC42 submitted 2024-12-04 cs.LG

Mahalakshmi Sabanayagam , Omar Al-Dabooni , Pascal Esser This is my paper · ORCID
classification cs.LG
keywords representationlearningcluster-specificembeddingstensorizedautoencodersvariationalcontrastiverestrictedBoltzmannmachinesunsupervisedclusteringdenoising
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 a downstream-agnostic definition of a good representation: when data have inherent clusters, the embedding function should be specific to each cluster rather than tuned to a single task. To make this practical, it introduces partial tensorization, which keeps most of the encoder shared and makes only a thin per-cluster layer cluster-specific, so the parameter cost grows slowly with the number of clusters. The authors instantiate this idea in autoencoders, variational autoencoders, contrastive learning, and restricted Boltzmann machines, and show on several datasets that the cluster-specific versions improve clustering and denoising over standard counterparts. The central claim is that the cluster-specific inductive bias, not extra model capacity, is what yields the better representations.

What carries the argument

The load-bearing mechanism is partial tensorization: the embedding function is factored into a shared network $g_\Omega$ followed by $k$ small cluster-specific maps $g_{\Psi_j}$, together with a cluster assignment matrix $S$ updated by k-means-style Lloyd steps. The shared part learns structures common to all clusters; the per-cluster part tunes the embedding to each cluster, so the method inherits the benefits of full tensorization without scaling the whole model with $k$. A k-means penalty $\lambda \|g_{\Psi_j}(g_\Omega(\tilde{x}_{i,j}))\|_2^2$ is added in the autoencoder setting to keep the latent space cluster-friendly.

What would settle it

Train PTAE on a dataset with known cluster structure, first with $k$ equal to the true number of clusters and then with $k$ larger than the true count; in the overestimated case, check whether the extra per-cluster embedding functions actually converge to copies of existing ones as the paper predicts, and whether clustering ARI and denoising MSE stay stable or degrade — either outcome would settle the claim.

Watch

Extended reading notes

Core claim

The paper's central claim is that a representation is good when it is specific to each inherent cluster in the data, and that this property can be learned jointly with cluster assignments through a partial tensorized objective: minimize $\sum_{j=1}^k \sum_{i=1}^n S_{j,i} \mathcal{L}(g_{\Psi_j}(g_\Omega(x_i)))$ subject to $S$ being a valid soft assignment matrix, where $g_\Omega$ is a shared encoder and each $g_{\Psi_j}$ is a small cluster-specific layer. For reconstruction-based models, the same pattern is applied to encoder and decoder with cluster-centered inputs $\tilde{x}_{i,j}=x_i-c_j$. Empirically, the partial tensorized autoencoder (PTAE) outperforms a standard autoencoder on clustering and denoising and matches the fully tensorized autoencoder while using far fewer parameters, because only the final layer is duplicated per cluster.

Load-bearing premise

The method requires the number of clusters $k$ to be known in advance, and every experiment sets $k$ to the true number of classes, so behavior under misspecified $k$ is only discussed heuristically.

Editorial extensions

If this is right

  • If the central claim holds, representation quality can be evaluated and improved without specifying a downstream task, because cluster specificity acts as a proxy for usefulness across tasks such as clustering and denoising.
  • Partial tensorization can be layered onto any representation learner with a small architectural change, so autoencoders, VAEs, contrastive models, and RBMs can all inherit cluster-specific embeddings.
  • New points are embedded by assigning them to the cluster whose embedding function gives the lowest loss, giving an explicit inference rule for unseen data.
  • Because the parameter overhead is confined to the per-cluster layer, the approach remains practical for larger numbers of clusters than fully tensorized versions.

Reading between the lines

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

  • A direct extension the paper leaves implicit is monitoring trained per-cluster layers for redundancy: pairwise distances between the $g_{\Psi_j}$ weight matrices could turn the heuristic claim about overestimated $k$ into a quantitative diagnostic.
  • The same shared-plus-per-cluster factorization could be applied at the output heads of a supervised network, effectively creating a lightweight mixture-of-experts that routes by latent cluster rather than by a hand-defined task.
  • Because the assignment matrix $S$ is updated by Lloyd steps, a mini-batch variant that assigns and updates on each batch could scale the method to streaming or very large datasets.
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 proposes "cluster specific representation learning," a downstream-agnostic formulation in which, given data with k assumed clusters, the learned representation is a concatenation of cluster-specific embedding functions applied to a shared representation. To control parameter growth, the authors introduce a partial tensorized variant (PTAE) that keeps a shared encoder and makes only a final layer cluster-specific, while still jointly optimizing a cluster assignment matrix S via an alternating scheme of gradient descent and Lloyd's steps. The idea is extended to Variational Autoencoders, contrastive losses, and Restricted Boltzmann Machines. The reported experiments compare PTAE with standard AEs and fully tensorized AEs (TAE) on clustering and denoising tasks, and qualitatively analyze VAE latent spaces. The central empirical claim is that PTAE outperforms standard AEs by a considerable margin and performs comparably to TAE with much lower model complexity.

Significance. If the central claim were rigorously established, the paper would make a useful contribution: it offers a principled, task-agnostic criterion for representation quality, and the partial-tensorization idea is a sensible way to control the added parameter cost of cluster-specific models. The paper also has a commendable breadth, showing that the same meta-formulation can be plugged into several representation-learning frameworks, and it provides code for reproduction. However, the current experimental evidence does not isolate the claimed mechanism: the PTAE objective differs from the AE baselines in two ways at once (a k-means penalty term and the joint clustering protocol), so the reported gains cannot yet be attributed to cluster specificity itself. The comparison also lacks error bars, statistical tests, and standard deep-clustering baselines. These are fixable issues, and the conceptual framework is worth pursuing, but the paper in its present form does not yet support its main quantitative claims.

major comments (4)
  1. [Section III-B, Figure 2 (left)] The clustering comparison is not on equal footing: for TAE and PTAE the reported ARI is computed from the assignment matrix S that is part of the joint optimization, while for standard AE baselines the ARI is obtained by running k-means on the learned embeddings. This asymmetry alone can explain part of the observed gap, because the tensorized methods receive direct gradient feedback on S. To support the claim that cluster-specific representations are intrinsically better, the authors should report k-means ARI on PTAE embeddings as well, and also run the tensorized assignment procedure on the standard AE embeddings, so that both methods are scored under identical clustering protocols.
  2. [Equation (3), Section III-A and Figure 2] The PTAE objective contains the extra term -lambda p(x~_{i,j}) with p(.) = ||g_{Psi_j}(g_Omega(x~_{i,j}))||^2, a k-means penalty that is absent from every AE baseline. Any improvement in clustering or denoising could therefore be caused by this regularizer alone, rather than by making the encoder cluster-specific. The authors should run an ablation with lambda = 0, and ideally also compare against a shared-encoder AE that uses the same k-means penalty without cluster-specific heads. This is load-bearing for the paper's central claim, and the absence of this ablation makes the quantitative evidence in Figure 2 inconclusive.
  3. [Section III-B, Figure 2 and Appendix VIII-B] The experimental evidence is presented without error bars, multiple restarts, or significance tests, and the main quantitative evaluation relies on two small toy datasets (parallel lines and 3D lines), one real low-dimensional dataset (penguin), and a 1000-sample subset of MNIST. Given that both the optimization and the clustering are sensitive to initialization and to the alternating updates, single-run point estimates are not sufficient to support the claimed 'considerable margin'. The authors should report means and variances over several random seeds, and ideally include standard deep clustering or deep embedding baselines (e.g., DEC or k-means-friendly autoencoders) to calibrate the magnitude of the reported ARI values.
  4. [Section VII and Appendix VIII-C] The discussion of an unknown number of clusters is purely heuristic. The claim that overestimating k makes 'the additional embedding functions simply learn copies of one of the cluster specific embedding functions' is not demonstrated anywhere in the paper; Appendix VIII-C only shows the underestimation case (six classes, two embedding functions). Since the method assumes k is known and all experiments set k to the true number of classes, the authors should either provide a diagnostic for choosing k, or empirically test the overestimation behavior to substantiate the claim in Section VII.
minor comments (6)
  1. [Section II, Optimization] There is a typo in 'Lyod's step' in the paragraph describing the alternating optimization; it should read 'Lloyd's step'.
  2. [Section III-B, Figure 2 left] The figure legend says 'k-means as benchmark' but the text describes the baseline as k-means on the original data; please make the protocol explicit in the caption.
  3. [Equation (6)] The KL divergence term has a mismatched parenthesis: it should be KL(q_{Psi_j,Omega}(z_j|x_i) || p_{Phi_j,Theta}(z_j)), not KL(q_{Psi_j,Omega}(z_j|x_i) || p_{Phi_j,Theta}(z_j).
  4. [Figure 3 caption] The caption repeats 'Top row: embedding of test samples' after the middle-row description; the third row should presumably be labeled differently (e.g., 'Bottom row').
  5. [Appendix VIII-B, Runtime] The text says 'for the datasets considered in Figure 2 and Figure 8 (top row) we provide the runtime comparison in Figure 8 (bottom row)', but Figure 8 does not have a top row of datasets; both rows are runtime plots. Please correct the cross-reference.
  6. [Section IV-A, Sampling] The reparameterization is written as z_{i,j} = mu_{i,j} + sigma^2_{i,j} * epsilon; this should be sigma_{i,j} (not sigma^2) for a standard Gaussian perturbation, or the notation for the variance should be clarified.

Circularity Check

1 steps flagged · score 4.0 of 10

Clustering evaluation uses the optimized assignment matrix S from Eq. (3), making the ARI comparison partially self-fulfilling; denoising and architecture claims remain independent.

  1. fitted input called prediction [Section III-A Eq. (3) and Section III-B 'Clustering' paragraph]
    "For evaluating clustering, we use the Adjusted Rand Index (ARI) values in [−1, 1] [13] between the true labels and the predictions. ... We compare the obtained clustering by i) k-means on the original data ii) k-means on the embedding obtained by standard AE and iii) the cluster assignment obtained by TAE and PTAE through S."

    The 'cluster assignment obtained by TAE and PTAE through S' is the same matrix S that PTAE minimizes in Eq. (3): min over {Phi_j,Psi_j}, Theta, Omega, S of sum_i sum_j S_j,i [ ||x~_i,j - x-hat_i,j||^2 - lambda p(x~_i,j) ], with S updated by Lloyd's step and cluster centers c_j recomputed from the current assignments. Thus the ARI reported for PTAE scores an optimized training variable (plus the k-means penalty) as if it were an independent clustering prediction from the learned representation. The AE baselines, by contrast, are scored by running k-means on their embeddings, so the comparison is structurally asymmetric and the clustering gain is partly forced by the objective rather than demonstrated as a property of the cluster-specific structure.

full rationale

The paper's core architectural contribution -- partial tensorization, splitting the encoder into a shared g_Omega and cluster-specific g_Psi_j (Eq. 1) -- is defined independently of the evaluation and is not circular. Citations to [9], which shares authors, are used as a baseline and for the initialization/Lloyd protocol, but they are not invoked as a uniqueness theorem or as the proof of PTAE's validity, so the self-citation is not load-bearing. The partial circularity is in the clustering experiment: the matrix S minimized in Eq. (3) (including the k-means penalty p and cluster centers recomputed from S) is the same object reported as the clustering prediction in Section III-B and scored with ARI, while AE baselines are scored by running k-means on their embeddings. This makes the clustering comparison structurally asymmetric and the clustering gain partly an artifact of the training objective; a lambda=0 ablation or a shared-encoder model with the same penalty would be needed to attribute the gain to cluster-specific structure. The denoising comparison, the model-complexity comparison, and the qualitative VAE/CL/RBM analyses do not share this loop, so the central claims retain independent content. Overall score 4.

Assumptions & free parameters 2 free parameters · 3 assumptions · 0 invented entities

The central claim relies on the assumption that k inherent clusters exist and that a single cluster-specific layer is sufficient. These are not derived or independently verified. The k-means penalty strength lambda is a hand-chosen hyperparameter. No new entities are postulated.

free parameters (2)
  • lambda (k-means penalty strength) = not specified
    Regularization weight in PTAE objective (Eq. 3), chosen by hand, not tuned across datasets.
  • number of clusters k = set to true number of classes C in experiments
    The method requires k as input; experiments set k=C, so the clustering performance is evaluated at the known-correct k.
assumptions (3)
  • domain assumption The data contains k inherent clusters
    Section II: 'we assume the data has k clusters'. The entire method depends on this.
  • ad hoc to paper A one-layer cluster-specific encoder is sufficient to capture cluster-specific structure
    Section II: 'we consider the setting where g_Psi_j only consists of one fully connected layer'. This architectural choice is not justified theoretically.
  • domain assumption Cluster assignment for new points by minimizing loss (Eq. 2) is correct
    Section II: inference rule assumes the loss is a good proxy for cluster membership.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Cluster Specific Representation Learning." pith.science (2026). https://pith.science/paper/NPUPSC42

@misc{pith2026241203471,
  author       = {Pith},
  title        = {Pith review of: Cluster Specific Representation Learning},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/NPUPSC42}},
  note         = {Machine review of arXiv:2412.03471}
}
read the original abstract

Representation learning aims to extract meaningful lower-dimensional embeddings from data, known as representations. Despite its widespread application, there is no established definition of a ``good'' representation. Typically, the representation quality is evaluated based on its performance in downstream tasks such as clustering, de-noising, etc. However, this task-specific approach has a limitation where a representation that performs well for one task may not necessarily be effective for another. This highlights the need for a more agnostic formulation, which is the focus of our work. We propose a downstream-agnostic formulation: when inherent clusters exist in the data, the representations should be specific to each cluster. Under this idea, we develop a meta-algorithm that jointly learns cluster-specific representations and cluster assignments. As our approach is easy to integrate with any representation learning framework, we demonstrate its effectiveness in various setups, including Autoencoders, Variational Autoencoders, Contrastive learning models, and Restricted Boltzmann Machines. We qualitatively compare our cluster-specific embeddings to standard embeddings and downstream tasks such as de-noising and clustering. While our method slightly increases runtime and parameters compared to the standard model, the experiments clearly show that it extracts the inherent cluster structures in the data, resulting in improved performance in relevant applications.

Figures

Figures reproduced from arXiv: 2412.03471 by the authors.

Figure 1
Figure 1. Representations obtained by a linear AE. [PITH_FULL_IMAGE:figures/full_fig_p001_1.png] view at source ↗
Figure 2
Figure 2. Performance of AE, TAE and PTAE. Left: clustering obtained by different models with k-means as benchmark. Plotted is ARI (higher is better). Center: de-noising of different models. MSE in log-scale (lower is better). Right: number of parameters in each considered model. idea is to map the data into a lower dimensional space using an encoder and then back into the original space using a decoder. In most cases, both t… view at source ↗
Figure 3
Figure 3. Top row: samples from the latent space. The very left plots shows samples from the latent space of the standard VAE. The following three plots show samples from the cluster specific latent spaces of the TVAE. Middle row: embedding of training samples. Embedding of training data-points using the standard VAE and TVAE. Plotted is the obtained mean as well as the contour lines of Gaussian model. Top row: embedding of t… view at source ↗
Figures from the paper (8 more)
Figure 4
Figure 4. Figure 4: Illustration of the embedding obtained from the (tensorized) contrastive [PITH_FULL_IMAGE:figures/full_fig_p005_4.png]
Figure 5
Figure 5. Figure 5: Reconstruction through TRBM. Left: true samples Middle/Right: reconstruction for class one and two. observe that for the specified class, the model manages to reconstruct the given images, while reconstructions of digits from the other class only result in random noise…
Figure 6
Figure 6. Figure 6: Model illustration. Top: reconstruction loss based partial tenzorised model. Bottom: tensorized model with loss on the obtained embeddings [PITH_FULL_IMAGE:figures/full_fig_p007_6.png]
Figure 8
Figure 8. Figure 8: Runtime per epoch comparison between AE, TAE and PTAE. [PITH_FULL_IMAGE:figures/full_fig_p008_8.png]
Figure 7
Figure 7. Figure 7: Additional datasets. Performance of AE, TAE and PTAE. [PITH_FULL_IMAGE:figures/full_fig_p008_7.png]
Figure 13
Figure 13. Figure 13: ACKNOWLEDGMENTS This work was done when Omar Al-Dabooni was a Bach￾elor’s student at the Technical University of Munich. The authors acknowledge the funding assistance from projects, specifically the German Research Foundation (Priority Pro￾gram SPP 2298, project GH 2…
Figure 12
Figure 12. Figure 12: Illustration of the samples and elastic transformation applied on the [PITH_FULL_IMAGE:figures/full_fig_p009_12.png]
Figure 11
Figure 11. Figure 11: Same setup as Figure 3, however for MNIST with sigmoid output [PITH_FULL_IMAGE:figures/full_fig_p009_11.png]

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

51 extracted references · 45 canonical work pages

  1. [9]

    Im- proved representation learning through tensorized autoencoders,

    P. Esser, S. Mukherjee, M. Sabanayagam, and D. Ghoshdastidar, “Im- proved representation learning through tensorized autoencoders,” in In- ternational Conference on Artificial Intelligence and Statistics . PMLR, 2023

    work page 2023

  2. [1]

    Extract- ing and composing robust features with denoising autoencoders,

    P. Vincent, H. Larochelle, Y . Bengio, and P.-A. Manzagol, “Extract- ing and composing robust features with denoising autoencoders,” in Proceedings of the 25th international conference on Machine learning , 2008

    work page 2008

  3. [2]

    Beyond a gaussian denoiser: Residual learning of deep cnn for image denoising,

    K. Zhang, W. Zuo, Y . Chen, D. Meng, and L. Zhang, “Beyond a gaussian denoiser: Residual learning of deep cnn for image denoising,” IEEE Transactions on Image Processing , 2017

    work page 2017

  4. [3]

    Topology and data,

    G. Carlsson, “Topology and data,” Bulletin of the American Mathemat- ical Society, 2009

    work page 2009

  5. [4]

    Topology-preserving deep image segmentation,

    A. Nigmetov, M. K. Biswas, K. Pogorelov, M. Riegler, P. Halvorsen, and S. Hicks, “Topology-preserving deep image segmentation,” in Proceed- ings of the European Conference on Computer Vision (ECCV) , 2020

    work page 2020

  6. [5]

    Nonlinear principal component analysis using autoas- sociative neural networks,

    M. A. Kramer, “Nonlinear principal component analysis using autoas- sociative neural networks,” AIChE Journal, vol. 37, no. 2, pp. 233–243, 1991

    work page 1991

  7. [6]

    The interpretation of interaction in contingency tables,

    E. H. Simpson, “The interpretation of interaction in contingency tables,” Journal of the Royal Statistical Society. Series B (Methodological), 1951

    work page 1951

  8. [7]

    Simpson’s paradox in real life,

    C. H. Wagner, “Simpson’s paradox in real life,” The American Statisti- cian, 1982

    work page 1982

Show all 51 references
  1. [8]

    Potential simpson’s paradox in multicenter study of intraperitoneal chemotherapy for ovarian cancer,

    G. B. Holt, “Potential simpson’s paradox in multicenter study of intraperitoneal chemotherapy for ovarian cancer,” Journal of Clinical Oncology, 2016

    2016

  2. [10]

    Auto-encoding variational bayes,

    D. P. Kingma and M. Welling, “Auto-encoding variational bayes,” arXiv preprint arXiv:1312.6114, 2013

    2013 arXiv

  3. [11]

    Information processing in dynamical systems: Foun- dations of harmony theory,

    P. Smolensky, “Information processing in dynamical systems: Foun- dations of harmony theory,” in Parallel distributed processing: Ex- plorations in the microstructure of cognition, Vol. 1: Foundations , Fig. 12. Illustration of the samples and elastic transformation applied on...

    1986

  4. [12]

    K-means++: The advantages of careful seeding,

    D. Arthur and S. Vassilvitskii, “K-means++: The advantages of careful seeding,” Proc. of the Annu. ACM-SIAM Symp. on Discrete Algorithms , 2007

    2007

  5. [13]

    Comparing partitions,

    L. Hubert and P. Arabie, “Comparing partitions,” Journal of Classifica- tion, 1985

    1985

  6. [14]

    Reducing the dimensionality of data with neural networks,

    G. E. Hinton and R. R. Salakhutdinov, “Reducing the dimensionality of data with neural networks,” Science, 2006

    2006

  7. [15]

    A review of image denoising algorithms, with a new one,

    A. Buades, B. Coll, and J.-M. Morel, “A review of image denoising algorithms, with a new one,” Multiscale modeling & simulation , 2005

    2005

  8. [16]

    Towards k- means-friendly spaces: Simultaneous deep learning and clustering,

    B. Yang, X. Fu, N. D. Sidiropoulos, and M. Hong, “Towards k- means-friendly spaces: Simultaneous deep learning and clustering,” in international conference on machine learning . PMLR, 2017

    2017

  9. [17]

    A survey on masked autoencoder for self-supervised learning in vision and beyond,

    C. Zhang, C. Zhang, J. Song, J. S. K. Yi, K. Zhang, and I. S. Kweon, “A survey on masked autoencoder for self-supervised learning in vision and beyond,” arXiv preprint arXiv:2208.00173 , 2022

    2022 arXiv

  10. [18]

    Towards k- means-friendly spaces: Simultaneous deep learning and clustering,

    B. Yang, X. Fu, N. D. Sidiropoulos, and M. Hong, “Towards k- means-friendly spaces: Simultaneous deep learning and clustering,” in Proceedings of the 34th International Conference on Machine Learning , 2017

    2017

  11. [19]

    Mnist handwritten digit database,

    Y . LeCun and C. Cortes, “Mnist handwritten digit database,” http://yann.lecun.com/exdb/mnist/, 2010

    2010

  12. [20]

    Ecological sexual di- morphism and environmental variability within a community of antarctic penguins (genus pygoscelis),

    K. B. Gorman, T. D. Williams, and W. R. Fraser, “Ecological sexual di- morphism and environmental variability within a community of antarctic penguins (genus pygoscelis),” PLOS ONE, 2014

    2014

  13. [21]

    Tutorial on variational autoencoders,

    C. Doersch, “Tutorial on variational autoencoders,” arXiv preprint arXiv:1606.05908, 2016

    2016 arXiv

  14. [22]

    An introduction to variational autoencoders,

    D. P. Kingma, M. Welling et al. , “An introduction to variational autoencoders,” Foundations and Trends® in Machine Learning , 2019

    2019

  15. [23]

    Generating sentences from a continuous space,

    S. R. Bowman, L. Vilnis, O. Vinyals, A. M. Dai, R. Jozefowicz, and S. Bengio, “Generating sentences from a continuous space,” in Proceedings of the 20th SIGNLL conference on computational natural language learning, 2015

    2015

  16. [24]

    Variational autoencoder based anomaly detection using reconstruction probability,

    J. An and S. Cho, “Variational autoencoder based anomaly detection using reconstruction probability,” Special Lecture on IE , 2015

    2015

  17. [25]

    beta-vae: Learning basic visual concepts with a constrained variational framework,

    I. Higgins, L. Matthey, A. Pal, C. Burgess, X. Glorot, M. Botvinick, S. Mohamed, and A. Lerchner, “beta-vae: Learning basic visual concepts with a constrained variational framework,” arXiv preprint arXiv:1611.02731, 2017

    2017 arXiv

  18. [26]

    Signature verification using a

    J. Bromley, I. Guyon, Y . LeCun, E. S ¨ackinger, and R. Shah, “Signature verification using a” siamese” time delay neural network,” Advances in neural information processing systems , 1993

    1993

  19. [27]

    Self-supervised representation learning: Introduction, advances, and challenges,

    L. Ericsson, H. Gouk, C. C. Loy, and T. M. Hospedales, “Self-supervised representation learning: Introduction, advances, and challenges,” IEEE Signal Processing Magazine , 2022

    2022

  20. [28]

    Self-supervised learning: Generative or contrastive,

    X. Liu, F. Zhang, Z. Hou, L. Mian, Z. Wang, J. Zhang, and J. Tang, “Self-supervised learning: Generative or contrastive,” IEEE transactions on knowledge and data engineering , 2021

    2021

  21. [29]

    Warpnet: Weakly su- 10 pervised matching for single-view reconstruction,

    A. Kanazawa, D. W. Jacobs, and M. Chandraker, “Warpnet: Weakly su- 10 pervised matching for single-view reconstruction,” in IEEE Conference on Computer Vision and Pattern Recognition , 2016

    2016

  22. [30]

    A simple framework for contrastive learning of visual representations,

    T. Chen, S. Kornblith, M. Norouzi, and G. Hinton, “A simple framework for contrastive learning of visual representations,” in International conference on machine learning , 2020

    2020

  23. [31]

    Self-supervised visual feature learning with deep neural networks: A survey,

    L. Jing and Y . Tian, “Self-supervised visual feature learning with deep neural networks: A survey,” IEEE Transactions on Pattern Analysis and Machine Intelligence, 2019

    2019

  24. [32]

    Self-supervised learning of pretext- invariant representations,

    I. Misra and L. v. d. Maaten, “Self-supervised learning of pretext- invariant representations,” in Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , 2020

    2020

  25. [33]

    BERT: pre-training of deep bidirectional transformers for language understanding,

    J. Devlin, M. Chang, K. Lee, and K. Toutanova, “BERT: pre-training of deep bidirectional transformers for language understanding,” in Proceedings of the 2019 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies,, 2019

    2019

  26. [34]

    Self-supervised learning of geometrically stable features through probabilistic introspec- tion,

    D. Novotny, S. Albanie, D. Larlus, and A. Vedaldi, “Self-supervised learning of geometrically stable features through probabilistic introspec- tion,” in IEEE Conference on Computer Vision and Pattern Recognition, 2018

    2018

  27. [35]

    Unsupervised repre- sentation learning by predicting image rotations,

    S. Gidaris, P. Singh, and N. Komodakis, “Unsupervised repre- sentation learning by predicting image rotations,” arXiv preprint arXiv:1803.07728, 2018

    2018 arXiv

  28. [36]

    A theoretical analysis of contrastive unsupervised representation learning,

    S. Arora, H. Khandeparkar, M. Khodak, O. Plevrakis, and N. Saunshi, “A theoretical analysis of contrastive unsupervised representation learning,” in International Conference on Machine Learning , 2019

    2019

  29. [37]

    Best practices for convolutional neural networks applied to visual document analysis,

    P. Simard, D. Steinkraus, and J. Platt, “Best practices for convolutional neural networks applied to visual document analysis,” in Seventh In- ternational Conference on Document Analysis and Recognition, 2003. Proceedings., 2003

    2003

  30. [38]

    Contrastive learning with hard negative samples,

    J. Robinson, C.-Y . Chuang, S. Sra, and S. Jegelka, “Contrastive learning with hard negative samples,” arXiv preprint arXiv:2010.04592 , 2021

    2010 arXiv

  31. [39]

    Supervised contrastive learn- ing,

    P. Khosla, P. Teterwak, C. Wang, A. Sarna, Y . Tian, P. Isola, A. Maschinot, C. Liu, and D. Krishnan, “Supervised contrastive learn- ing,” in Advances in Neural Information Processing Systems , 2020

    2020

  32. [40]

    Supporting supervised contrastive learning via contrastive label disambiguation,

    W. Zhang, G. Meng, J. Zhou, D. Tao, and J. Xie, “Supporting supervised contrastive learning via contrastive label disambiguation,” in Proceed- ings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, 2021

    2021

  33. [41]

    Visualizing data using t-sne,

    L. van der Maaten and G. Hinton, “Visualizing data using t-sne,” Journal of Machine Learning Research , 2008. [Online]. Available: http://jmlr.org/papers/v9/vandermaaten08a.html

    2008

  34. [42]

    Near-optimal comparison based clustering,

    M. Perrot, P. Esser, and D. Ghoshdastidar, “Near-optimal comparison based clustering,” Advances in Neural Information Processing Systems , 2020

    2020

  35. [43]

    Semi-supervised learning,

    O. Chapelle, B. Scholkopf, and A. Zien, “Semi-supervised learning,” IEEE Transactions on Neural Networks , 2009

    2009

  36. [44]

    Deep adaptive image clustering,

    J. Chang, L. Wang, G. Meng, S. Xiang, and C. Pan, “Deep adaptive image clustering,” Proceedings of the IEEE International Conference on Computer Vision (ICCV) , 2017

    2017

  37. [45]

    Fischer and C

    A. Fischer and C. Igel, Training restricted Boltzmann machines: An introduction. Springer, 2014

    2014

  38. [46]

    Deep learning for anomaly detection in time-series data: Review, analysis, and guidelines,

    D. Choi, B. Jeong, S. T. Jeong, and K. S. Lee, “Deep learning for anomaly detection in time-series data: Review, analysis, and guidelines,” in IEEE Access, 2019

    2019

  39. [47]

    A fast learning algorithm for deep belief nets,

    G. E. Hinton, S. Osindero, and Y .-W. Teh, “A fast learning algorithm for deep belief nets,” Neural computation, 2006

    2006

  40. [48]

    The role of occam’s razor in knowledge discovery,

    P. Domingos, “The role of occam’s razor in knowledge discovery,” Data Mining and Knowledge Discovery , vol. 3, no. 4, pp. 409–425, 1999

    1999

  41. [49]

    Vapnik, Statistical Learning Theory

    V . Vapnik, Statistical Learning Theory . Wiley-Interscience, 1998

    1998

  42. [50]

    The use of multiple measurements in taxonomic prob- lems,

    R. A. Fisher, “The use of multiple measurements in taxonomic prob- lems,” Annals of Human Genetics , 1936

    1936

  43. [51]

    Fashion-mnist: a novel image dataset for benchmarking machine learning algorithms,

    H. Xiao, K. Rasul, and R. V ollgraf, “Fashion-mnist: a novel image dataset for benchmarking machine learning algorithms,” arXiv, 2017

    2017

Pith tools

Reviewed August 11, 2026 · model on record in the stance chip above.