Pith. sign in

REVIEW 4 major objections 5 minor 16 references

Clustering via Self-Supervised Diffusion

T0 review · 4 major / 5 minor · reviewed 2026-08-06 · deepseek-v4-flash

Pith's one-line read CLUDI repurposes a diffusion model as a clustering engine and beats all compared unsupervised classifiers.

desk verdict New clustering method that couples diffusion-sampled assignment embeddings with self-distillation; the diffusion component is plausible but the paper never tests whether it actually drives the gains. read the letter →

arxiv 2507.04283 v2 pith:KW4HLYOC submitted 2025-07-06 cs.AI cs.CV

classification cs.AIcs.CV MSC 62H3068T07
keywords diffusionmodelsclusteringself-supervisedlearningunsupervisedclassificationteacher-studentdistillationvisiontransformersDINOfeaturesDDIM
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

CLUDI is, by its own account, the first framework to turn a diffusion model into a clustering engine. It takes feature vectors from a pre-trained Vision Transformer (DINO) and trains a diffusion model to map random noise into "assignment embeddings" that encode a probability distribution over K clusters for each image. Because reverse sampling is stochastic, one image yields many different plausible assignments; a teacher branch generates these, and a student branch is trained to reproduce them from corrupted inputs. At inference the model averages over many sampled assignments to produce its final prediction. The paper reports that this scheme beats prior self-supervised clustering methods on every benchmark it tests (CIFAR-10, STL-10, Oxford-IIIT Pets, Oxford 102 Flower, Caltech-101, and the ImageNet-50/100/200 subsets), a result that matters because clustering underlies label-free tasks in segmentation, anomaly detection, and bioinformatics.

What carries the argument

The load-bearing object is a feature-conditioned diffusion model acting as a stochastic cluster-assignment sampler. Starting from $z_T \sim \mathcal{N}(0, F^2 I_d)$, a 25-step stochastic DDIM reverse process (Eqs. 7-9) denoises to an assignment embedding $z_0$, which a logit projection $L$ and tempered softmax convert into cluster probabilities. Three pieces carry the argument: (i) the Monte Carlo averaging of $B$ sampled probability vectors (Eq. 3), which stabilizes predictions; (ii) the teacher-student self-distillation scheme with SimSiam-style shared weights, where the student must predict both the denoised embedding $z_0$ (MSE loss) and the cluster probabilities $u$ (cross-entropy with a uniform-prior regularizer drawn from Self-Classifier); and (iii) the auxiliary embedding matrix $E$ that maps probabilities back into the embedding space, playing the role of the predictor network in BYOL and SimSiam and enabling the embedding targets in Eq. 13. The sqrt noise schedule and Min-SNR-$\gamma$ weighting control how different noise levels contribute to training.

What would settle it

Train CLUDI's identical student and losses but replace the diffusion teacher with a non-diffusion stochastic teacher that emits the same number $B$ of diverse assignment embeddings per image (for example, noise-injected MLP outputs with the same target normalization); if it matches CLUDI's accuracy on ImageNet-100 validation, the diffusion mechanism is not the load-bearing ingredient. A second check: compare CLUDI at $B=1$ against the averaged $B>1$ prediction — if single-sample accuracy already matches the averaged accuracy, the Monte Carlo averaging in Eq. 3 contributes nothing.

Watch

Extended reading notes

Core claim

The central claim is that the stochasticity of a diffusion model can serve as a structured data-augmentation mechanism for cluster assignment, not just for image generation. Concretely, CLUDI defines the cluster-posterior $p_\theta(k|x)$ as a Monte Carlo average over $B$ samples from a data-conditioned diffusion model, $p_\theta(k|x) \approx \frac{1}{B}\sum_{i=1}^B p(k|z_0^i)$, where each $z_0^i \in \mathbb{R}^d$ is an assignment embedding obtained by stochastic DDIM reverse sampling starting from pure Gaussian noise. The teacher's diverse assignments become training targets for a student network, and training combines an MSE loss on denoised embeddings with a cross-entropy loss on cluster probabilities that is regularized by a uniform prior over the mini-batch to prevent collapse. The paper argues that this yields the best reported NMI, ACC, and ARI on all tested benchmarks, outperforming SCAN, ProPos, TSP, TEMI, and Self-Classifier variants, and that averaging over many samples makes predictions on heavily corrupted inputs nearly as accurate as on clean inputs.

Load-bearing premise

The load-bearing premise is that stochastic diversity from the diffusion sampler itself — 25 DDIM steps starting from pure noise — is what makes the teacher's assignment targets informative and accurate, yet the paper never tests a non-diffusion teacher, and its own ablations show results hinge on tuning the noise scale $F^2$ and embedding dimension $d$.

Editorial extensions

If this is right

  • On every setting the paper reports — CIFAR-10, STL-10, Oxford-IIIT Pets, Oxford 102 Flower, Caltech-101, and the ImageNet-50/100/200 subsets — CLUDI reports the highest NMI, ACC, and ARI among the compared self-supervised clustering methods, with its largest margins on the ImageNet subsets.
  • Accuracy on inputs corrupted by feature dropout and Gaussian noise approaches clean-input accuracy as the number of averaged assignment samples $B$ grows (Figure 2), so the method's robustness comes from sampling diversity rather than from extra augmentation pipelines or neighbor mining.
  • Because CLUDI works on frozen pre-trained features and needs no multi-head selection or pseudo-label refinement, the clustering stage is comparatively lightweight, inheriting the efficiency of pre-trained-feature pipelines while adding only the diffusion head.
  • The reported gains depend on the method's specific design choices — the $E$ matrix predictor, the uniform-prior classification loss, and the Min-SNR weighting — which together prevent the trivial solutions that earlier deep clustering methods struggled with.

Reading between the lines

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

  • If diffusion noise is the active ingredient, the same teacher-student scheme should transfer beyond images: any frozen embedding space (text, bioinformatics, graphs) could be clustered with a small diffusion head, which would be a cheap experiment to run.
  • A sharper test the authors leave implicit: replace the diffusion teacher with a non-diffusion stochastic sampler, such as an MLP emitting equally diverse noise-injected assignment embeddings with the same targets and losses; comparable accuracy would mean the sampling mechanism, not diffusion as such, is doing the work.
  • The sampling count $B$ acts as an inference-time compute knob — a deployment could run at low $B$ where latency matters and upgrade to high $B$ where accuracy matters, without retraining.
  • The paper's limitation section reports degradation as the number of clusters $K$ grows; a natural stress test is whether hierarchical or multi-scale diffusion with $K$-dependent noise schedules preserves cluster separability at large $K$, the direction the authors point to in their conclusion.
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 introduces CLUDI, a self-supervised clustering method that operates on pre-trained DINO ViT features. The core idea is to use a diffusion model as a teacher: conditioned on a feature vector x, the teacher runs a stochastic DDIM sampling procedure (25 training steps, 100 inference steps) from Gaussian noise to produce multiple assignment embeddings z0, which are converted to cluster probabilities via a trainable logit layer. The student, fed noisy/augmented versions of the input and of the teacher's embeddings, is trained to predict both the embeddings and the cluster probabilities under an MSE loss and a regularized cross-entropy loss with a uniform-prior term. At inference, the probabilities from B sampled embeddings are averaged. Experiments on CIFAR-10, STL-10, ImageNet subsets, and several fine-grained datasets report state-of-the-art clustering accuracy, with ablations over the loss weight λ, the noise scale F², the embedding dimension d, and the number of averaged samples B.

Significance. If the reported results hold, CLUDI would be a novel and effective application of diffusion models to clustering, and the paper's central claim that stochastic diffusion-based assignment generation improves unsupervised classification would be of clear interest to the deep clustering community. The paper is also commendable for providing code, for comparing against a frozen-feature Self-Classifier baseline on identical DINO features, and for running several hyperparameter ablations. However, the significance is currently qualified by the absence of a control that isolates the diffusion component from the other self-distillation and regularization ingredients, and by the lack of error bars and non-neural baselines; these omissions make the state-of-the-art claim and the attribution of the gain to diffusion uncertain.

major comments (4)
  1. [Section 5.1, 5.4 and Section 6] The paper claims that stochastic diffusion-based sampling is what drives performance, but no ablation removes or replaces the diffusion teacher while keeping all other components fixed. The only non-diffusion comparison, Self-Classifier*, differs from CLUDI in many simultaneous ways: it has no embedding MSE loss (Eq. 18), no student feature/embedding noise (Section 5.2), no E-projection (Eq. 13), and no B-averaging at inference. The ablations in Figures S1–S3 and Figure 2 vary λ, F², d, and B, but never replace the teacher with a deterministic or single-step denoiser. Therefore the reported gains could in principle come entirely from the two losses and the uniform-prior regularization applied to any stochastic embedding generator, with the 25-step diffusion process acting only as an expensive source of randomness. A direct control is needed: keep the student and losses identical, but replace the 25-step DDIM teacher with, for example, a single-step denoiser that maps (x, Gaussian noise) to an embedding, or with a deterministic MLP; if accuracy is unchanged, the load-bearing attribution to diffusion is not established.
  2. [Tables 1 and 2, Section 6] The main results are reported as single numbers with no error bars or number of seeds. Since several claimed wins are small (e.g., CIFAR-10 NMI 89.6 vs. TEMI 88.6, ImageNet-200 ACC 73.7 vs. 73.2), single runs cannot support a state-of-the-art claim. Additionally, there is no k-means or Gaussian-mixture baseline on the same DINO features; without such baselines, the reader cannot tell how much of the improvement comes from the clustering model rather than from the quality of the pre-trained features. Please report mean±standard deviation over at least three seeds for all tables and add k-means/GMM on the same frozen DINO features as lower-bound references.
  3. [Section 3 and Section 4, Eqs. (7)–(10)] The DDIM sampling equations are not self-consistent as written. Section 3 defines α_t := 1−β_t and ar α_t as the cumulative product, but Eq. (8) uses α_s and α_t in the standard DDIM update, which requires the cumulative product (usually denoted ar α_s and ar α_t), while Eq. (10) correctly uses ar α_t. As written, the variance terms and the noise prediction in Eq. (8) are inconsistent with the forward process in Eq. (2). Please correct the notation (either redefine α_t as the cumulative product in Section 4 or replace α_s, α_t with ar α_s, ar α_t in Eq. (8)), so that the method is reproducible from the text alone.
  4. [Section 5.4, Eqs. (21)–(23)] The derivation of the regularized cross-entropy loss contains a garbled line. Eq. (23) reads 'p(k|\hat z) = p(k)p(k|\hat z)/p(k) = (NB/K) p(k|\hat z) Σ ...', which is tautological once p(k) is canceled. The intended expression should presumably define p(\hat z|k) via a column softmax analogous to Eq. (21) and then apply Bayes to obtain p(k|\hat z). As printed, the posterior probability formula is not well-defined and the loss in Eq. (24) cannot be reconstructed. Please rewrite Eqs. (21)–(23) with clear, correct definitions of the likelihood and posterior terms.
minor comments (5)
  1. [Section 6, Experimental Setup] The paper should state explicitly that all hyperparameters (d, F², λ) were selected on the ImageNet-100 validation set and then fixed for other datasets, and clarify whether the numbers in Tables 1–2 are test-set results or validation-set results for ImageNet-100.
  2. [Figure 2 caption] The caption says accuracy 'becomes similar as the number of data samples grows', but the x-axis appears to be B (number of averaged samples), not the number of data samples; please rephrase.
  3. [Section 5.4, Eq. (26)] The Min-SNR weighting as written uses max(SNR, γ)/(SNR+1). Please verify this against the cited source (Hang et al., 2023), which is commonly stated with a min(SNR, γ) factor; either the formula is a variant or the citation needs clarification.
  4. [Sections 5.1 and 6] The teacher uses 25 DDIM steps during training while all reported results use 100 inference steps; please state explicitly whether this difference is intentional and whether the student is trained only on targets produced by the 25-step teacher.
  5. [Section 5.1, Eq. (13)] The embedding matrix E is introduced but not explicitly defined as a learnable parameter; Figure 4 indicates it is trained, so the text should state its trainability and where its gradients flow.

Circularity Check

0 steps flagged · score 2.0 of 10

No significant circularity: CLUDI's derivation is self-contained; the self-distillation loop is explicit and standard, and the claims are grounded in external benchmarks.

full rationale

The paper's teacher-student setup is self-referential by design: the teacher and student share weights with a stop-gradient, and the target embeddings in Eq. 13 are derived from the teacher's own softmax outputs. However, this is explicitly described as self-distillation in Section 5, following BYOL/SimSiam/DINO, and is not presented as an independent ground truth. The load-bearing claim of state-of-the-art clustering is evaluated against external ground-truth labels on seven benchmarks, so the empirical result is not forced by construction. Hyperparameters d, F^2, and lambda are tuned on a validation split and then fixed, which is standard practice rather than a fitted input being renamed a prediction. The absence of a non-diffusion teacher control weakens the causal attribution of the gains to diffusion, but that is an experimental-support limitation, not circularity. Self-citations (Pakman et al. 2020; Chelly et al. 2025; Ronen et al. 2022) appear only in related work and are not load-bearing. No uniqueness theorem or ansatz is imported from the authors' prior work; the adopted diffusion components are drawn from external references (Ho et al., Song et al., Li et al., Salimans & Ho, Hang et al.).

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

The central claim rests on the quality of frozen DINO features, on 25-step stochastic DDIM sampling producing informative assignment targets, on the cited non-collapse property of the regularized cross-entropy loss, and on several tuned hyperparameters (F^2, d, lambda). No new physical or mathematical entities are introduced.

free parameters (6)
  • F^2 = 25.0
    Diffusion noise rescaling factor; tuned on ImageNet-100 validation, with values 4, 9, 16, 25, 36, 49 tested in Appendix C; adopted for K>=100 datasets.
  • d = 64 (K>=100), 32 (K<100)
    Embedding dimension; tuned on ImageNet-100 validation; scanning shown in Figure 7 and Appendix C.
  • lambda = 50
    Weight on classification loss; tuned on ImageNet-100 validation; ablations in Figure 6 and Appendix B.
  • B = not reported in main text
    Number of stochastic samples averaged at inference; Figure 2 shows accuracy vs B, but the final value used in Tables 1-2 is not stated.
  • tau, tau_col = not reported
    Softmax temperatures in the classification head and column softmax; values are not listed in the paper.
  • Min-SNR gamma = 5
    Fixed threshold in loss weighting; cited from Hang et al. 2023, not tuned here.
assumptions (4)
  • domain assumption DINO ViT features pre-trained on ImageNet provide sufficiently informative representations for clustering the target datasets.
    Used as frozen input x throughout Sections 4-6; no end-to-end feature learning.
  • domain assumption The 25-step DDIM sampling that maps pure noise z_T ~ N(0,F^2 I) to assignment embeddings z_0 produces diverse, useful cluster targets for the student.
    Assumed in Sections 5.1 and 5.3; no ablation shows that this diversity, rather than the regularization losses, drives performance.
  • standard math The uniform-prior cross-entropy loss of Amrani et al. prevents collapsed solutions as claimed.
    Adopted in Equations 21-24; the non-collapse property is cited, not re-derived here.
  • standard math DDPM and DDIM forward and reverse processes (Equations 1-11) correctly model the conditional distribution of assignment embeddings.
    Background diffusion theory from Ho et al. 2020 and Song et al. 2021b.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Clustering via Self-Supervised Diffusion." pith.science (2026). https://pith.science/paper/KW4HLYOC

@misc{pith2026250704283,
  author       = {Pith},
  title        = {Pith review of: Clustering via Self-Supervised Diffusion},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/KW4HLYOC}},
  note         = {Machine review of arXiv:2507.04283}
}
read the original abstract

Diffusion models, widely recognized for their success in generative tasks, have not yet been applied to clustering. We introduce Clustering via Diffusion (CLUDI), a self-supervised framework that combines the generative power of diffusion models with pre-trained Vision Transformer features to achieve robust and accurate clustering. CLUDI is trained via a teacher-student paradigm: the teacher uses stochastic diffusion-based sampling to produce diverse cluster assignments, which the student refines into stable predictions. This stochasticity acts as a novel data augmentation strategy, enabling CLUDI to uncover intricate structures in high-dimensional data. Extensive evaluations on challenging datasets demonstrate that CLUDI achieves state-of-the-art performance in unsupervised classification, setting new benchmarks in clustering robustness and adaptability to complex data distributions. Our code is available at https://github.com/BGU-CS-VIL/CLUDI.

Figures

Figures reproduced from arXiv: 2507.04283 by the authors.

Figure 1
Figure 1. Overview of the CLUDI Framework at Infer￾ence. Images pass through a pre-trained Vision Transformer to obtain their feature representations. Multiple random vectors are sampled from a Gaussian distribution. A diffu￾sion model, conditioned on the features, refines the random vectors into class assignment embeddings. Each refined embedding corresponds to a candidate cluster probability vector. By averaging multiple su… view at source ↗
Figure 2
Figure 2. Classification accuracy for clean and aug￾mented inputs. The model classification accuracy on aug￾mented data (feature dropout plus Gaussian noise) becomes similar to that of clean data as the number of data samples grows. Results from ImageNet 100 validation data. Stan￾dard deviation based on 10 repetitions. Here z˜θ(zt, x, t) : R d → R d is a network that predicts z0, and is trained by minimizing Ez0,zt,t,x h w(t)… view at source ↗
Figure 3
Figure 3. sqrt noise scheduling. The noise grows faster near t = 0, reflecting reduced sensitivity to noise at early timesteps, and grows gradually slower at later times. where the teacher and student networks share weights, un￾like BYOL and DINO, where the teacher is updated via an exponential moving average of the student. A key distinction of CLUDI is that, for each data point x, the student has two different learning targ… view at source ↗
Figures from the paper (4 more)
Figure 4
Figure 4. Figure 4: Overview of CLUDI’s training phase. Given a set of image features x, the teacher model generates denoised assignment embeddings z˜0 which are used to create two targets for the student: (i) clustering probabilities u and (ii) assignment embeddings z0, obtained from u v…
Figure 5
Figure 5. Figure 5: Left: t-SNE visualization of the assignment embedding space of ImageNet 50 demonstrating the model’s ability to organize data points into well-separated clusters. Right: Examples of correctly classified images (cropping, color alterations, etc), our augmentations act d…
Figure 7
Figure 7. Figure 7: Embedding dimension selection d. Each point in the curves shows the maximum validation accuracy on ImageNet-100 achieved during training. Acknowledgments This work was supported in part by the Lynn and William Frankel Center at BGU CS, by Israel Science Foundation Pers…
Figure 6
Figure 6. Figure 6: Ablation Study on the Lcls weight λ. Each point in the curves shows the maximum validation accuracy on ImageNet-100 achieved during training [PITH_FULL_IMAGE:figures/full_fig_p009_6.png]

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

16 extracted references · 3 canonical work pages

  1. [3]

    H., Doucet, A., Strudel, R., Dyer, C., Durkan, C., et al

    Dieleman, S., Sartran, L., Roshannai, A., Savinov, N., Ganin, Y ., Richemond, P. H., Doucet, A., Strudel, R., Dyer, C., Durkan, C., et al. Continuous diffusion for categorical data. arXiv preprint arXiv:2211.15089,

  2. [7]

    Empowering diffusion models on the embedding space for text generation

    Gao, Z., Guo, J., Tan, X., Zhu, Y ., Zhang, F., Bian, J., and Xu, L. Empowering diffusion models on the embedding space for text generation. In Proceedings of the 2024 Conference of the North American Chapter of the Associ- ation for Computational Linguistics: Human Language Technologies (Volume 1: Long Papers), pp. 4664–4683,

  3. [9]

    Varia- tional deep embedding: An unsupervised and generative approach to clustering

    Jiang, Z., Zheng, Y ., Tan, H., Tang, B., and Zhou, H. Varia- tional deep embedding: An unsupervised and generative approach to clustering. arXiv preprint arXiv:1611.05148,

  4. [10]

    Understanding diffusion models: A unified perspec- tive

    Luo, C. Understanding diffusion models: A unified perspec- tive. arXiv preprint arXiv:2208.11970,

  5. [11]

    Step- by-step diffusion: An elementary tutorial

    Nakkiran, P., Bradley, A., Zhou, H., and Advani, M. Step- by-step diffusion: An elementary tutorial. arXiv preprint arXiv:2406.08929,

  6. [12]

    and Zisserman, A

    Nilsback, M.-E. and Zisserman, A. Automated flower clas- sification over a large number of classes. In 2008 Sixth Indian conference on computer vision, graphics & image processing, pp. 722–729. IEEE,

  7. [14]

    E., Diaconu, C.-D., Markou, S., Shysheya, A., Foong, A

    Turner, R. E., Diaconu, C.-D., Markou, S., Shysheya, A., Foong, A. Y ., and Mlodozeniec, B. Denoising diffusion probabilistic models in six simple steps. arXiv preprint arXiv:2402.04384,

  8. [15]

    S., and Tian, Y

    Wang, X., Chen, X., Du, S. S., and Tian, Y . Towards de- mystifying representation learning with non-contrastive self-supervision. arXiv preprint arXiv:2110.04947,

Show all 16 references
  1. [16]

    and Zhang, N

    Zhou, X. and Zhang, N. L. Deep clustering with fea- tures from self-supervised pretraining. arXiv preprint arXiv:2207.13364,

  2. [2011]

    Imagenet: A large-scale hierarchical image database

    Deng, J., Dong, W., Socher, R., Li, L.-J., Li, K., and Fei-Fei, L. Imagenet: A large-scale hierarchical image database. In 2009 IEEE conference on computer vision and pattern recognition, pp. 248–255. Ieee,

  3. [2014]

    Learning generative visual models from few training examples: An incremen- tal bayesian approach tested on 101 object categories

    Fei-Fei, L., Fergus, R., and Perona, P. Learning generative visual models from few training examples: An incremen- tal bayesian approach tested on 101 object categories. In 2004 conference on computer vision and pattern recogni- tion workshop, pp. 178–178. IEEE,

  4. [2019]

    An image is worth 16x16 words: Trans- formers for image recognition at scale

    Dosovitskiy, A. An image is worth 16x16 words: Trans- formers for image recognition at scale. arXiv preprint arXiv:2010.11929,

  5. [2020]

    M., Vedaldi, A., Zisserman, A., and Jawahar, C

    Parkhi, O. M., Vedaldi, A., Zisserman, A., and Jawahar, C. Cats and dogs. In 2012 IEEE conference on computer vision and pattern recognition , pp. 3498–3505. IEEE,

  6. [2021]

    Chan, S. H. Tutorial on diffusion models for imaging and vision. arXiv preprint arXiv:2403.18103,

  7. [2022]

    Distributed mcmc inference in dirichlet process mixture models using julia

    Dinari, O., Yu, A., Freifeld, O., and Fisher, J. Distributed mcmc inference in dirichlet process mixture models using julia. In 2019 19th IEEE/ACM International Symposium on Cluster, Cloud and Grid Computing (CCGRID), pp. 518–525. IEEE,

  8. [2024]

    Diffuseq: Sequence to sequence text generation with diffusion mod- els

    Gong, S., Li, M., Feng, J., Wu, Z., and Kong, L. Diffuseq: Sequence to sequence text generation with diffusion mod- els. arXiv preprint arXiv:2210.08933,

Pith tools

Reviewed August 6, 2026 · model on record in the stance chip above.