Pith. sign in

REVIEW 4 major objections 5 minor 113 references

SGN: A Similarity-based Generative Network for Data Generation under Distribution Shift

T0 review · 4 major / 5 minor · reviewed 2026-08-01 · deepseek-v4-flash

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

Pith's one-line read A generative model trained once on labeled source data can synthesize useful target-domain samples under distribution shift by learning a latent space organized by zero-one label similarities, with no parameter updates at generation time.

desk verdict Practical niche and a plausible trick — zero-one similarity latent space plus target-rep mixing — but the paper never checks that a decoder trained on scalar rescalings of single encodings can decode fresh linear combinations of several encodings, and the headline image comparison hands SGN labeled target examples the CGMs never see. read the letter →

arxiv 2607.18072 v1 pith:XAZKFWK4 submitted 2026-07-20 cs.LG cs.AI

Jiaqi Zhu , Xincheng Chen , Yuncheng Wu , Zhaojing Luo , Beng Chin Ooi This is my paper
classification cs.LGcs.AI
keywords distributionshiftdataaugmentationgenerativemodellatentspacezero-onesimilarityclassconsistencyautoencodertransferablegeneration
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

This paper proposes SGN, a generative network trained once on labeled source data and then applied to new target domains without any parameter updates. The key claim is that instead of learning a source data distribution, SGN learns a latent space in which the inner product between two samples' special representations equals a zero-one label similarity: one if labels match, zero otherwise. At generation time, a small labeled set from the target domain is encoded, same-label representations are linearly combined, and the decoder maps the result back to a new target-domain sample. If the claim holds, a single model can produce useful target-domain augmentations under prior or covariate shift, improving downstream classifiers and regressors without per-target fine-tuning.

What carries the argument

The central object is the zero-one similarity matrix S and the latent-space constraint ⟨x_lat_S_i, x_lat_S_j⟩ = 1 if y_i = y_j, 0 otherwise. This constraint makes same-label latent representations share an orthogonal basis and different-label representations orthogonal, so the latent space encodes class geometry independently of the input distribution. An autoencoder is trained with a regularization loss for this inner-product condition and a reconstruction loss; at generation, the trained encoder maps a few labeled target samples into this space, and a linear combination of equal-label latent representations is fed to the decoder to synthesize a new sample.

What would settle it

Compute the pairwise inner-product matrix of encoded special latent representations for a held-out target representative set and compare it with the label-based zero-one matrix; a large entrywise mismatch for a shifted domain directly contradicts the class-consistency premise. A second check: if generated samples do not improve or actually degrade downstream accuracy relative to training on the target set alone, the practical claim fails.

Watch

Extended reading notes

Core claim

SGN replaces distribution learning with similarity-structure learning. The training objective forces the encoder to produce latent representations whose special forms satisfy the zero-one similarity requirement: the inner product of two special latent representations is 1 for same-label pairs and 0 for different-label pairs, while a decoder reconstructs original inputs from colinear latent perturbations. The paper argues that this label-induced geometry is domain-stable: because the similarity requirement is expressed purely through inner products, applying the same encoder to target-domain representatives and linearly mixing equal-label representations inherits target-specific appearance wh

Load-bearing premise

The load-bearing premise is that the encoder and decoder trained only on source data continue to satisfy the zero-one similarity condition on target-domain representatives, so that linear combinations of target encodings decode into plausible, class-consistent target samples.

Editorial extensions

If this is right

  • A model trained once on one domain can be reused for multiple shifted target domains without storing per-target parameters.
  • Generated target-domain samples can be used as augmentation to improve downstream classifiers and regressors where source-trained conditional GANs and VAEs fail or hurt.
  • The dimension lower bound D ≥ B + K − 2Q gives a concrete architectural guideline for the latent space in terms of batch size and number of labels.
  • Generation reduces to solving a small linear system in latent space, so no iterative optimization is needed at inference time.
  • Because p(y) and p(x|y) are sampled from the target representative set, the method addresses both prior probability shift and covariate shift.

Reading between the lines

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

  • The latent-space recipe is effectively a learned, label-conditioned interpolation: if the encoder is locally linear across target examples, the method can be seen as an implicit manifold mixup with a class-consistency certificate only to the extent the similarity constraint transfers.
  • The zero-one similarity is blind to within-class structure; a harder test would be fine-grained or hierarchical labels, where a graded similarity might be necessary.
  • A natural extension is to use unlabeled target data with pseudo-labels or clustering to build the representative set, removing the need for labeled target samples.
  • If the transfer assumption breaks—for example, under a large rotation or style shift where the encoder sends same-label target points apart—the paper's mechanism offers no fallback; monitoring target-pair inner products would be a practical safeguard.
Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, and a circularity audit.

Referee Report

4 major / 5 minor

Summary. The paper proposes SGN, an autoencoder-based generative framework trained once on labeled source data. Training enforces a zero-one similarity structure in latent space: the inner product of two special latent representations is 1 for same-label pairs and 0 otherwise, while a reconstruction term preserves the mapping to the original space. At generation time, given a small labeled representative set from a target domain, SGN encodes target examples, constructs new latent codes (in the experiments, by linearly mixing same-class target encodings), and decodes them to synthesize target-domain samples for augmentation. The paper also derives a latent-dimension lower bound, presents a so-called convergence analysis (Theorems 3.4–3.5), and reports experiments on six image datasets and five tabular datasets against conditional GANs and tabular generative baselines.

Significance. If validated, the proposed reusable, fine-tuning-free approach to target-guided data augmentation would be practically attractive, and the use of label-induced similarity geometry is an interesting alternative to explicit class-conditioning. The paper is broad in scope, with experiments spanning image and tabular domains, and the image results in Table 1 show large nominal improvements. The strengths are the simple training objective and the clear separation between source training and target-time generation. However, the central mechanism is not adequately verified: the decoder is trained on an input distribution different from the one used at generation time, the main comparisons give SGN access to labeled target representatives while baselines do not see such information, and the theoretical convergence analysis is largely existential rather than about training dynamics. These issues prevent the current evidence from supporting the paper's central claim.

major comments (4)
  1. [§3.2 & §3.4, Eq. (3) and Algorithm 1] The decoder is trained only on inputs of the form c·xlat_i/||xlat_i|| (scalar rescalings of a single latent code), but at generation Step 3 constructs xlat_new by linearly mixing multiple same-class target encodings and Step 4 feeds this vector to the decoder. A linear combination of non-collinear latent codes is not a scalar multiple of any training input, so the decoder has never been trained to invert such vectors. The paper states in Section 3.4 that these latent representations 'inherently consist of the distribution information of the target domain' and that decoding yields target-domain samples, but no experiment, architecture argument, or bound verifies the decoder's behavior on this input distribution. This mismatch is independent of distribution shift and directly threatens the method's validity. Additionally, Algorithm 1's claim that xlat_new 'satisfies Equation (1)' is not ju
  2. [§4.2, Table 1 and §4.3, Tables 2–3] The headline comparison is not controlled. SGN receives a labeled representative set XRep from the target domain and uses it to determine p(y), p(x|y) and to construct xlat_new; the four image baselines and CTGAN/TVAE are trained on the source and generate from the source distribution, without access to any target examples. The large gains in Table 1 may therefore reflect the extra target-side information rather than the proposed similarity mechanism. To support the claim, the authors should include target-informed baselines (e.g., input-space or latent Mixup on target representatives, few-shot adapted generators, or an oracle that trains a conditional generator on target representatives) and report significance/variance.
  3. [§3.3, Theorems 3.4–3.5] This is not a convergence analysis. Theorem 3.4 restates the global optimality condition X^lat_S*(X^lat_S*)^T = S. Theorem 3.5 assumes the original data space is discrete/enumerable and proves that for any real symmetric S_W there exists a matrix X with XX^T = S_W; this is an existence statement about matrix factorization, not about the training dynamics of the proposed loss. It does not show that gradient-based optimization of Eq. (3) converges to such a solution. The paper should either retitle this section as a realizability/dimension analysis or provide an actual convergence guarantee.
  4. [§3.4, last paragraph] The central transfer claim—that target latent representations 'inherently consist of the distribution information of the target domain' and that decoding their mixture yields target-domain samples—is asserted without proof and without verification. No metric is reported for whether generated samples are closer to the target domain than source-trained baselines, and no class-consistency check on generated samples is provided. A transfer bound on the encoder/decoder under covariate shift, or at least quantitative domain-alignment and class-consistency diagnostics, is needed.
minor comments (5)
  1. [§4.1] The text promises a discussion of choosing f(·) 'for different types of datasets in Section 4', but Section 4 only lists the identities (NeurSpec, NormSpec) and does not explain why identity is preferred for images and normalization for tabular data.
  2. [§4.2] The text refers to 'RcACGAN' while Table 1 uses 'ReACGAN'; please harmonize the naming.
  3. [Figure 2] The textual description of the visualization is confusing ('the third generated image ... i.e., the fifth image in the fifth row'); please clarify the referencing and mark the relevant images in the figure.
  4. [§4.3, Tables 2–3] The tabular results report only point estimates. Include standard deviations or confidence intervals, especially for comparisons that are close (e.g., Adult1→Adult0 with MLP).
  5. [§3.2, footnote 1] The footnote assumes data in a batch have different special latent representations. For the normalized f used by NormSpec, two same-label special representations at the optimum must be identical (unit vectors with inner product 1), so the 'worst-case' assumption is inconsistent with that realization.

Circularity Check

2 steps flagged · score 6.0 of 10

The target-distribution result is imported through the representative inputs and the optimality theorem restates the loss; remaining empirical claims are tested, so partial circularity.

  1. self definitional [Section 3.3, Theorem 3.4 (page 6)]
    "Recall that any design of L(r1)i,j and L(r2)i is proposed to satisfy the aforementioned two requirements, and thus the global optimality essentially satisfies the two requirements, as stated in theorem 3.4. Theorem 3.4. Xlat_S∗ (Xlat_S∗) T = S, where Xlat_S∗ represents the optimal solution of special latent representations Xlat_S."

    The 'optimal solution' is defined by minimizing Eq. (3)-(4), where L(r1) penalizes |<f(xlat_i), f(xlat_j)> - 1| for same labels and |<f(xlat_i), f(xlat_j)>| for different labels. Setting this loss to zero is exactly Eq. (1), i.e., Xlat_S(Xlat_S)^T = S. The theorem therefore restates the training objective as a conclusion; it supplies no independent realizability or convergence result. Any subsequent claim that SGN 'finds' this latent structure inherits the target rather than deriving it.

  2. fitted input called prediction [Section 3.4, Algorithm 1 and final paragraph (page 7)]
    "Next, we construct a latent representation xlat_new that satisfies Equation (1)... In the experiments, we utilize a simple but effective way that linearly mixes the data points in Xequal... we can sample the label value ak and the Xequal dataset according to p(y) and p(x|y = ak), respectively... These latent representations inherently consist of the distribution information of the target domain. With the mixture of the latent representations, we can make the generated data satisfy the target domain distribution."

    Algorithm 1 obtains target p(y) and p(x|y) directly from XRep, encodes those target representatives, and forms xlat_new as a linear mix of same-class target encodings. The decoder maps this mixture back to data space. Thus the 'target-specific characteristics' and the target distribution of the generated samples are inherited from the input representative set by construction; the claim that this 'tackle[s] the prior probability shift and covariate shift' is a restatement of feeding the target distribution into the generator rather than a prediction or derivation from the source-trained model. The only part not fixed by construction—whether the decoder is valid on such mixtures—is asserted without proof.

full rationale

The paper's self-citations (Zhu et al. 2023/2025/2026) are contextual and not load-bearing, so they do not raise the score. Theorems 3.4 and 3.5 do not provide an independent derivation: 3.4 is the global minimum of Eq. (3)-(4) by definition, and 3.5 proves only that a symmetric Gram matrix has a factorization. The main reduction is in Section 3.4: the target domain's p(y) and p(x|y) are taken from XRep and linearly mixed in latent space, so the generated samples inherit target characteristics by construction. The downstream classification/regression experiments on held-out target data provide some independent evidence of utility, which is why the paper is not wholly circular. But the central 'distribution-shift solution' is partially baked into Algorithm 1, and the decoder's behavior on linear combinations of multiple target encodings—an input class never trained (training uses c·xlat_i/||xlat_i||, Eq. 3's reconstruction term)—is asserted rather than derived. Score 6.

Assumptions & free parameters 5 free parameters · 6 assumptions · 1 invented entities

The central claim rests on unverified transfer of the similarity geometry from source to target, plus several manually chosen hyperparameters (η, C, D, Nk, f). The theoretical results are consequences of the definition of the similarity loss, not independent evidence.

free parameters (5)
  • η (loss trade-off) = not reported in main text (ablation in Appendix D)
    Balances similarity regularization L(r1) and reconstruction L(r2) in Eq. (3); must be chosen per dataset/run.
  • C (latent length perturbation) = not reported
    Uniform sampling width around latent norm for decoder input; set by hand; no selection rule.
  • D (latent dimension) = not reported in main text
    Must obey D ≥ B+K−2Q (Theorem 3.3) but actual value is chosen by practitioner; affects capacity and similarity satisfaction.
  • Nk (number of selected target reps per label) = 200 per label for images; tabular not stated
    Algorithm 1's Xselect size; generation behavior and output diversity depend on it.
  • f(·) post-processing function = identity (NeurSpec) or normalized (NormSpec)
    Post-processing function choice changes training objective and results; paper reports both and prefers NeurSpec without a selection principle.
assumptions (6)
  • domain assumption Source-trained encoder preserves the zero-one similarity geometry on target data (Eq. (1) holds for target encodings).
    Training in §3.2 only uses source data; §3.4 encodes target representatives and assumes their inner products still reflect same/different labels despite shift. No bound or experiment verifies this.
  • domain assumption Decoder trained on source can map novel linear combinations of target encodings to plausible target-domain samples.
    Generation decodes c·x_lat_new where x_lat_new is a mix of target latent vectors; decoder has never seen target-domain data or such combinations, and generalization is assumed.
  • domain assumption Availability of a small labeled representative set XRep from the target domain with correct labels.
    Algorithm 1 requires labeled target samples; this is a precondition for use, possibly costly/unavailable.
  • ad hoc to paper The original data space is discrete and enumerable (Theorem 3.5).
    Introduced solely to make the existence-of-representation argument for convergence; not true for continuous image/tabular data.
  • standard math Any real symmetric similarity matrix S admits a factorization S=XX^T.
    Used in proof of Theorem 3.5; true for PSD matrices, but convergence of neural-network training does not follow from existence.
  • domain assumption Gradient-based training reaches the global optimum satisfying both losses.
    No convergence proof for non-convex autoencoder objective; Theorem 3.5 only asserts existence of a matrix, not that SGD finds it.
invented entities (1)
  • special latent representation x_lat_S
    purpose: Carries label-induced zero-one similarity via inner products; central object used for generation.
    The paper postulates a latent representation whose pairwise inner products equal the label similarity matrix; no independent falsifiable handle outside its own loss.

how reviews work

0 comments
Cite this review

Pith. "Pith review of SGN: A Similarity-based Generative Network for Data Generation under Distribution Shift." pith.science (2026). https://pith.science/paper/XAZKFWK4

@misc{pith2026260718072,
  author       = {Pith},
  title        = {Pith review of: SGN: A Similarity-based Generative Network for Data Generation under Distribution Shift},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/XAZKFWK4}},
  note         = {Machine review of arXiv:2607.18072}
}
read the original abstract

Generative models trained on a source domain often produce samples that are poorly aligned with shifted target domains, limiting their effectiveness for target-domain data augmentation. Although target-specific adaptation can reduce this mismatch, it typically requires additional optimization and domain-specific parameters. We propose a Similarity-based Generative Network (SGN), a reusable framework that is trained once on labeled source data and applied to new target domains without parameter updates. SGN learns a latent space structured by label-induced pairwise similarities while preserving reconstructive information through an encoder-decoder architecture. At generation time, a small labeled representative set from the target domain is encoded and combined in the learned latent space, allowing the generated samples to inherit target-specific characteristics while maintaining class consistency. We further analyze the realizability and dimensionality requirements of the proposed similarity structure. Experiments on image and tabular datasets demonstrate the effectiveness of SGN for target-guided data augmentation under source-to-target distribution shifts.

Figures

Figures reproduced from arXiv: 2607.18072 by the authors.

Figure 1
Figure 1. An illustration of the SGN paradigm, which consist [PITH_FULL_IMAGE:figures/full_fig_p004_1.png] view at source ↗
Figure 2
Figure 2. Visualization results of the RetinaMNIST and Tiss [PITH_FULL_IMAGE:figures/full_fig_p009_2.png] view at source ↗

Discussion (0). Sign in to comment.

Reference graph

Works this paper leans on

113 extracted references · 9 linked inside Pith

  1. [1]

    SIGKDD , pages =

    Ron Kohavi , title =. SIGKDD , pages =

  2. [2]

    Towards Concise Models of Grid Stability , booktitle =

    Vadim Arzamasov and Klemens B. Towards Concise Models of Grid Stability , booktitle =

  3. [3]

    Multiclass classification of dry beans using computer vision and machine learning techniques , journal =

    Murat Koklu and Ilker Ali. Multiclass classification of dry beans using computer vision and machine learning techniques , journal =

  4. [4]

    Claudio De Stefano and Francesco Fontanella and Marilena Maniaci and Alessandra Scotto di Freca , title =

  5. [5]

    Challenges in benchmarking stream learning algorithms with real-world data , journal =

    Vin. Challenges in benchmarking stream learning algorithms with real-world data , journal =

  6. [6]

    Yunpei Jia and Jie Zhang and Shiguang Shan and Xilin Chen , title =

  7. [7]

    Single-Domain Generalization in Medical Image Segmentation via Test-Time Adaptation from Shape Dictionary , booktitle =

    Quande Liu and Cheng Chen and Qi Dou and Pheng. Single-Domain Generalization in Medical Image Segmentation via Test-Time Adaptation from Shape Dictionary , booktitle =

  8. [8]

    Xinjie Fan and Qifei Wang and Junjie Ke and Feng Yang and Boqing Gong and Mingyuan Zhou , title =

Show all 113 references
  1. [9]

    Zijian Wang and Yadan Luo and Ruihong Qiu and Zi Huang and Mahsa Baktashmotlagh , title =

  2. [10]

    Fengchun Qiao and Long Zhao and Xi Peng , title =

  3. [11]

    Jindong Wang and Cuiling Lan and Chang Liu and Yidong Ouyang and Tao Qin , title =

  4. [12]

    Modeling Tabular data using Conditional

    Lei Xu and Maria Skoularidou and Alfredo Cuesta. Modeling Tabular data using Conditional. NeurIPS , pages =

  5. [13]

    Conditional

    Hou, Liang and Cao, Qi and Shen, Huawei and Pan, Siyuan and Li, Xiaoshuang and Cheng, Xueqi , booktitle =. Conditional. 2022 , volume =

    2022

  6. [14]

    Introduction to statistical machine learning , author=

  7. [15]

    arXiv preprint arXiv:1905.10887 , year=

    Classification accuracy score for conditional generative models , author=. arXiv preprint arXiv:1905.10887 , year=

    1905 arXiv

  8. [16]

    arXiv:2110.14795 , year=

    MedMNIST v2: A Large-Scale Lightweight Benchmark for 2D and 3D Biomedical Image Classification , author=. arXiv:2110.14795 , year=

    arXiv

  9. [17]

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

    Codella, Noel and Rotemberg, Veronica and Tschandl, Philipp and Celebi, M Emre and Dusza, Stephen and others , journal=. Skin lesion analysis toward melanoma detection 2018: A challenge hosted by the international skin imaging collaboration (

    2018

  10. [18]

    2020 , author =

    A dataset of microscopic peripheral blood cell images for development of automatic recognition systems , journal =. 2020 , author =

    2020

  11. [19]

    NeurIPS Workshop , year=

    Reading digits in natural images with unsupervised feature learning , author=. NeurIPS Workshop , year=

  12. [20]

    arXiv preprint arXiv:1708.07747 , year=

    Fashion-mnist: a novel image dataset for benchmarking machine learning algorithms , author=. arXiv preprint arXiv:1708.07747 , year=

    arXiv

  13. [21]

    Proceedings of the IEEE , volume=

    Gradient-based learning applied to document recognition , author=. Proceedings of the IEEE , volume=

  14. [22]

    Understanding the Limitations of Conditional Generative Models , booktitle =

    Ethan Fetaya and J. Understanding the Limitations of Conditional Generative Models , booktitle =

  15. [23]

    SIGKDD , pages=

    Spectral Clustering of Attributed Multi-relational Graphs , author=. SIGKDD , pages=

  16. [24]

    Neurocomputing , volume=

    A survey of deep neural network architectures and their applications , author=. Neurocomputing , volume=

  17. [25]

    ICCV , pages=

    Singan: Learning a generative model from a single natural image , author=. ICCV , pages=

  18. [26]

    ICCV , pages=

    Dualgan: Unsupervised dual learning for image-to-image translation , author=. ICCV , pages=

  19. [27]

    arXiv preprint arXiv:1701.00160 , year=

    Nips 2016 tutorial: Generative adversarial networks , author=. arXiv preprint arXiv:1701.00160 , year=

    2016 arXiv

  20. [28]

    arXiv preprint arXiv:2112.07041 , year=

    Survey of Generative Methods for Social Media Analysis , author=. arXiv preprint arXiv:2112.07041 , year=

    arXiv

  21. [29]

    ACM Computing Surveys , volume=

    A survey on adversarial recommender systems: from attack/defense strategies to generative adversarial networks , author=. ACM Computing Surveys , volume=

  22. [30]

    ISCV , pages=

    Deep generative models: Survey , author=. ISCV , pages=. 2018 , organization=

    2018

  23. [31]

    AI EDAM , volume=

    Challenges in the industrial implementation of generative design systems: An exploratory study , author=. AI EDAM , volume=

  24. [32]

    ImageNet Classification with Deep Convolutional Neural Networks , volume =

    Krizhevsky, Alex and Sutskever, Ilya and Hinton, Geoffrey E , booktitle =. ImageNet Classification with Deep Convolutional Neural Networks , volume =

  25. [33]

    and Karypis, G

    Steinbach, M. and Karypis, G. and Kumar, V. , booktitle =

  26. [34]

    Journal of the American Statistical association , volume=

    Objective criteria for the evaluation of clustering methods , author=. Journal of the American Statistical association , volume=

  27. [35]

    NIPS , pages=

    Fundamental limitations of spectral clustering , author=. NIPS , pages=. 2007 , organization=

    2007

  28. [36]

    Self-Tuning Spectral Clustering , booktitle =

    Lihi Zelnik. Self-Tuning Spectral Clustering , booktitle =

  29. [37]

    Cohen , title =

    Frank Lin and William W. Cohen , title =. ICML , pages =

  30. [38]

    2012 , school=

    Scalable methods for graph-based unsupervised and semi-supervised learning , author=. 2012 , school=

    2012

  31. [39]

    Applied intelligence , volume=

    Deflation-based power iteration clustering , author=. Applied intelligence , volume=

  32. [40]

    WWW , pages=

    Rosc: Robust spectral clustering on multi-scale data , author=. WWW , pages=

  33. [41]

    SIGKDD , pages=

    CAST: A Correlation-based Adaptive Spectral Clustering Algorithm on Multi-scale Data , author=. SIGKDD , pages=

  34. [42]

    IEEE Transactions on pattern analysis and machine intelligence , volume=

    Normalized cuts and image segmentation , author=. IEEE Transactions on pattern analysis and machine intelligence , volume=

  35. [43]

    NIPS , pages=

    On spectral clustering: Analysis and an algorithm , author=. NIPS , pages=

  36. [44]

    ICDM , pages=

    Diverse power iteration embeddings and its applications , author=. ICDM , pages=. 2014 , organization=

    2014

  37. [45]

    SIGKDD , pages=

    Fuse: Full spectral clustering , author=. SIGKDD , pages=

  38. [46]

    ICCV , pages=

    Correlation adaptive subspace segmentation by trace lasso , author=. ICCV , pages=

  39. [47]

    CVPR , pages=

    Smooth representation clustering , author=. CVPR , pages=

  40. [48]

    The Journal of Machine Learning Research , volume=

    Information theoretic measures for clusterings comparison: Variants, properties, normalization and correction for chance , author=. The Journal of Machine Learning Research , volume=

  41. [49]

    , author=

    A probabilistic classification system for predicting the cellular localization sites of proteins. , author=. Ismb , volume=

  42. [50]

    ICML , pages=

    Spectral clustering for multi-type relational data , author=. ICML , pages=

  43. [51]

    Pattern Recognition , volume=

    Enabling scalable spectral clustering for image segmentation , author=. Pattern Recognition , volume=

  44. [52]

    IEEE Transactions on Geoscience and Remote Sensing , volume=

    Spectral clustering ensemble applied to SAR image segmentation , author=. IEEE Transactions on Geoscience and Remote Sensing , volume=

  45. [53]

    Neurocomputing , volume=

    Spectral clustering of high-dimensional data exploiting sparse representation vectors , author=. Neurocomputing , volume=

  46. [54]

    Budhaditya Deb and Guoqing Zheng and Milad Shokouhi and Ahmed Hassan Awadallah , title =

  47. [55]

    CVPR , pages=

    Plug & play generative networks: Conditional iterative generation of images in latent space , author=. CVPR , pages=

  48. [56]

    Instance-Conditioned

    Arantxa Casanova and Marl. Instance-Conditioned. CoRR , volume =. 2021 , eprinttype =

    2021

  49. [57]

    CVPR , pages=

    Deep residual learning for image recognition , author=. CVPR , pages=

  50. [58]

    NIPS , volume=

    Faster r-cnn: Towards real-time object detection with region proposal networks , author=. NIPS , volume=

  51. [59]

    ICML , pages=

    Convolutional sequence to sequence learning , author=. ICML , pages=

  52. [60]

    2021 , isbn =

    Dai, Enyan and Shu, Kai and Sun, Yiwei and Wang, Suhang , title =. 2021 , isbn =

    2021

  53. [61]

    NIPS , volume=

    Generative adversarial nets , author=. NIPS , volume=

  54. [62]

    Kingma and Max Welling , title =

    Diederik P. Kingma and Max Welling , title =

  55. [63]

    arXiv preprint arXiv:1411.1784 , year=

    Conditional generative adversarial nets , author=. arXiv preprint arXiv:1411.1784 , year=

    arXiv

  56. [64]

    NIPS , volume=

    Learning structured output representation using deep conditional generative models , author=. NIPS , volume=

  57. [65]

    Xing , title =

    Zhiting Hu and Zichao Yang and Xiaodan Liang and Ruslan Salakhutdinov and Eric P. Xing , title =

  58. [66]

    Kai Shu and Suhang Wang and Thai Le and Dongwon Lee and Huan Liu , title =

  59. [67]

    Augustus Odena and Christopher Olah and Jonathon Shlens , title =

  60. [68]

    2022 , issn =

    Federated synthetic data generation with differential privacy , journal =. 2022 , issn =

    2022

  61. [69]

    arXiv preprint arXiv:2108.07927 , year=

    Fed-TGAN: Federated Learning Framework for Synthesizing Tabular Data , author=. arXiv preprint arXiv:2108.07927 , year=

    arXiv

  62. [70]

    Mirko Palato , title =

  63. [71]

    ICML , pages=

    Auxiliary deep generative models , author=. ICML , pages=

  64. [72]

    ICML , pages=

    Variational inference with normalizing flows , author=. ICML , pages=

  65. [73]

    ICCV , pages=

    Sym-parameterized dynamic inference for mixed-domain image translation , author=. ICCV , pages=

  66. [74]

    CVPR , pages=

    Learning pyramid-context encoder network for high-quality image inpainting , author=. CVPR , pages=

  67. [75]

    WACV , pages=

    Chromagan: Adversarial picture colorization with semantic class distribution , author=. WACV , pages=

  68. [76]

    Dataset shift in machine learning , author=

  69. [77]

    , author=

    Discriminative learning under covariate shift. , author=. Journal of Machine Learning Research , volume=

  70. [78]

    IEEE TKDE , volume=

    Learning from imbalanced data , author=. IEEE TKDE , volume=

  71. [79]

    Cell , volume=

    Identifying medical diagnoses and treatable diseases by image-based deep learning , author=. Cell , volume=

  72. [80]

    Cytometry Part A , volume=

    In situ classification of cell types in human kidney tissue using 3D nuclear staining , author=. Cytometry Part A , volume=

  73. [81]

    The 2nd diabetic retinopathy – grading and image quality estimation challenge , year =

    DeepDRiD. The 2nd diabetic retinopathy – grading and image quality estimation challenge , year =

  74. [82]

    NeurIPS , year=

    Contragan: Contrastive learning for conditional image generation , author=. NeurIPS , year=

  75. [83]

    NeurIPS , volume=

    Rebooting ACGAN: Auxiliary Classifier GANs with Stable Training , author=. NeurIPS , volume=

  76. [84]

    Takeru Miyato and Masanori Koyama , title =

  77. [85]

    NeurIPS , volume=

    Domain invariant representation learning with domain density transformations , author=. NeurIPS , volume=

  78. [86]

    IEEE Transactions on Biomedical Engineering , year=

    Domain adaptation for medical image analysis: a survey , author=. IEEE Transactions on Biomedical Engineering , year=

  79. [87]

    Kaiming He and Xiangyu Zhang and Shaoqing Ren and Jian Sun , title =

  80. [88]

    Clinical data sharing using Generative Adversarial Networks , volume =

    Ayyoubzadeh, Seyed and Ayyoubzadeh, Seyed Mehdi and Esmaeili, Marzieh , year =. Clinical data sharing using Generative Adversarial Networks , volume =. Connected Health , doi =

  81. [89]

    CoRR , volume =

    Hritik Bansal and Aditya Grover , title =. CoRR , volume =

  82. [90]

    CoRR , volume =

    Han Xiao and Kashif Rasul and Roland Vollgraf , title =. CoRR , volume =

  83. [91]

    Reading Digits in Natural Images with Unsupervised Feature Learning , author =

  84. [92]

    Diffusion Models:

    Ling Yang and Zhilong Zhang and Yang Song and Shenda Hong and Runsheng Xu and Yue Zhao and Wentao Zhang and Bin Cui and Ming. Diffusion Models:. 2024 , url =. doi:10.1145/3626235 , timestamp =

    2024 doi

  85. [93]

    Denoising Diffusion Probabilistic Models , booktitle =

    Jonathan Ho and Ajay Jain and Pieter Abbeel , editor =. Denoising Diffusion Probabilistic Models , booktitle =. 2020 , url =

    2020

  86. [95]

    Artificial Intelligence Review , year=

    Generative anomaly detection: a comprehensive review of modeling principles, advances, and future opportunities , author=. Artificial Intelligence Review , year=

  87. [96]

    IEEE Transactions on Pattern Analysis and Machine Intelligence , year=

    Catching every ripple: Enhanced anomaly awareness via dynamic concept adaptation , author=. IEEE Transactions on Pattern Analysis and Machine Intelligence , year=

  88. [97]

    Proceedings of the VLDB Endowment , volume=

    METER: A Dynamic Concept Adaptation Framework for Online Anomaly Detection , author=. Proceedings of the VLDB Endowment , volume=. 2023 , publisher=

    2023

  89. [98]

    International Conference on Machine Learning , pages=

    In-Context Adaptation to Concept Drift for Learned Database Operations , author=. International Conference on Machine Learning , pages=. 2025 , organization=

    2025

  90. [99]

    Conditional GANs with Auxiliary Discriminative Classifier , booktitle =

    Liang Hou and Qi Cao and Huawei Shen and Siyuan Pan and Xiaoshuang Li and Xueqi Cheng , editor =. Conditional GANs with Auxiliary Discriminative Classifier , booktitle =. 2022 , url =

    2022

  91. [100]

    2023 , url =

    William Peebles and Saining Xie , title =. 2023 , url =. doi:10.1109/ICCV51070.2023.00387 , timestamp =

    2023

  92. [101]

    Xirui Li and Charles Herrmann and Kelvin C. K. Chan and Yinxiao Li and Deqing Sun and Chao Ma and Ming. A Simple Approach to Unifying Diffusion-based Conditional Generation , booktitle =. 2025 , url =

    2025

  93. [102]

    Weber , title =

    Seyedmorteza Sadat and Manuel Kansy and Otmar Hilliges and Romann M. Weber , title =. The Thirteenth International Conference on Learning Representations,. 2025 , url =

    2025

  94. [103]

    MineGAN: Effective Knowledge Transfer From GANs to Target Domains With Few Images , booktitle =

    Yaxing Wang and Abel Gonzalez. MineGAN: Effective Knowledge Transfer From GANs to Target Domains With Few Images , booktitle =. 2020 , url =. doi:10.1109/CVPR42600.2020.00935 , timestamp =

    2020

  95. [104]

    International Conference on Learning Representations , year =

    Do We Need All the Synthetic Data? Targeted Image Augmentation via Diffusion Models , author =. International Conference on Learning Representations , year =

  96. [105]

    Manifold Mixup: Better Representations by Interpolating Hidden States , booktitle =

    Vikas Verma and Alex Lamb and Christopher Beckham and Amir Najafi and Ioannis Mitliagkas and David Lopez. Manifold Mixup: Better Representations by Interpolating Hidden States , booktitle =. 2019 , url =

    2019

  97. [106]

    mixup: Beyond Empirical Risk Minimization , booktitle =

    Hongyi Zhang and Moustapha Ciss. mixup: Beyond Empirical Risk Minimization , booktitle =. 2018 , url =

    2018

  98. [107]

    CoRR , volume =

    Yara Bahram and Melodie Desbos and Mohammadhadi Shateri and Eric Granger , title =. CoRR , volume =. 2025 , url =. doi:10.48550/ARXIV.2511.18281 , eprinttype =. 2511.18281 , timestamp =

    2025 doi

  99. [108]

    CoRR , volume =

    Jonathan Ho and Tim Salimans , title =. CoRR , volume =. 2022 , url =. doi:10.48550/ARXIV.2207.12598 , eprinttype =. 2207.12598 , timestamp =

    2022 doi arXiv

  100. [109]

    Diffusion Models Beat GANs on Image Synthesis , booktitle =

    Prafulla Dhariwal and Alexander Quinn Nichol , editor =. Diffusion Models Beat GANs on Image Synthesis , booktitle =. 2021 , url =

    2021

  101. [110]

    CoRR , volume =

    Jingyuan Zhu and Huimin Ma and Jiansheng Chen and Jian Yuan , title =. CoRR , volume =. 2022 , url =. doi:10.48550/ARXIV.2211.03264 , eprinttype =. 2211.03264 , timestamp =

    2022 doi arXiv

  102. [111]

    The Thirteenth International Conference on Learning Representations,

    Teng Hu and Jiangning Zhang and Ran Yi and Hongrui Huang and Yabiao Wang and Lizhuang Ma , title =. The Thirteenth International Conference on Learning Representations,. 2025 , url =

    2025

  103. [112]

    The Thirteenth International Conference on Learning Representations,

    Jincheng Zhong and Xiangcheng Zhang and Jianmin Wang and Mingsheng Long , title =. The Thirteenth International Conference on Learning Representations,. 2025 , url =

    2025

  104. [113]

    The Twelfth International Conference on Learning Representations,

    Brandon Trabucco and Kyle Doherty and Max Gurinas and Ruslan Salakhutdinov , title =. The Twelfth International Conference on Learning Representations,. 2024 , url =

    2024

  105. [114]

    2025 , url =

    Yanghao Wang and Long Chen , title =. 2025 , url =. doi:10.1109/CVPR52734.2025.02380 , timestamp =

    2025

Pith tools

Reviewed August 1, 2026 · model on record in the stance chip above.