pith. machine review for the scientific record. sign in

arxiv: 2605.08519 · v1 · submitted 2026-05-08 · 💻 cs.LG

Recognition: 2 theorem links

· Lean Theorem

SeBA: Semi-supervised few-shot learning via Separated-at-Birth Alignment for tabular data

Kacper Jurek , Wojciech Batko , Marek \'Smieja , Marcin Przewi\k{e}\'zlikowski
Authors on Pith no claims yet

Pith reviewed 2026-05-12 01:26 UTC · model grok-4.3

classification 💻 cs.LG
keywords semi-supervised few-shot learningtabular datajoint embeddingrepresentation alignmentnearest-neighbor correspondenceview separation
0
0 comments X

The pith

SeBA generates an output space improving feature-label relationships in semi-supervised few-shot tabular learning by splitting data into two views and aligning representations to nearest-neighbor matches without augmentations.

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The paper proposes Separated-at-Birth Alignment (SeBA) to handle semi-supervised few-shot learning on tabular data where standard augmentations risk distorting semantics. It splits each sample into two independent yet complementary views and aligns the learned representations of one view so they mirror the nearest-neighbor structure found in the second view. Theoretical analysis and experiments show this alignment produces an output space with tighter feature-label connections. The method reaches state-of-the-art accuracy on most benchmark datasets drawn from medicine, finance, and science.

Core claim

SeBA is a joint-embedding framework that separates tabular inputs into two complementary views and aligns the representation of one view to the nearest-neighbor correspondences present in the other view, thereby constructing an output space whose feature-label relationship is measurably stronger than those obtained by augmentation-based self-supervised baselines.

What carries the argument

Separated-at-Birth Alignment (SeBA), a joint-embedding procedure that partitions each tabular instance into two independent views and forces one view's embeddings to reproduce the nearest-neighbor geometry observed in the second view.

Load-bearing premise

Tabular data admits a split into two independent complementary views that preserves semantics, and mirroring nearest-neighbor structure across views reliably strengthens the feature-label relationship in the resulting output space.

What would settle it

A direct measurement on a held-out tabular benchmark showing that the post-alignment output space exhibits no higher feature-label correlation (or downstream accuracy) than the best augmentation-based semi-supervised baseline.

Figures

Figures reproduced from arXiv: 2605.08519 by Kacper Jurek, Marcin Przewi\k{e}\'zlikowski, Marek \'Smieja, Wojciech Batko.

Figure 1.1
Figure 1.1. Figure 1.1: Typical Semi-supervised Few-shot Learning (SS-FSL) approaches [PITH_FULL_IMAGE:figures/full_fig_p002_1_1.png] view at source ↗
Figure 3.1
Figure 3.1. Figure 3.1: Representation learning via Separated-at-Birth Alignment (SeBA). In each minibatch, we separate [PITH_FULL_IMAGE:figures/full_fig_p004_3_1.png] view at source ↗
Figure 4.1
Figure 4.1. Figure 4.1: Summary of performance ranks of each analyzed method across few-shot classification tasks. SeBA is, on average, the best-ranking approach [PITH_FULL_IMAGE:figures/full_fig_p006_4_1.png] view at source ↗
Figure 4.2
Figure 4.2. Figure 4.2: Analysis of SeBA on the Optdigits dataset: (a) the fraction of same-class samples among the k nearest [PITH_FULL_IMAGE:figures/full_fig_p007_4_2.png] view at source ↗
Figure 4.3
Figure 4.3. Figure 4.3: Ablation of the design aspects of SeBA ( [PITH_FULL_IMAGE:figures/full_fig_p008_4_3.png] view at source ↗
read the original abstract

Learning from scarce labeled data with a larger pool of unlabeled samples, known as semi-supervised few-shot learning (SS-FSL), remains critical for applications involving tabular data in domains like medicine, finance, and science. The existing SS-FSL methods often rely on self-supervised learning (SSL) frameworks developed for vision or language, which assume the availability of a natural form of data augmentations. For tabular data, defining meaningful augmentations is non-trivial and can easily distort semantics, limiting the effectiveness of conventional SSL. In this work, we rethink SSL for tabular data and propose Separated-at-Birth Alignment (SeBA), a joint-embedding framework for SS-FSL that eliminates the dependence on augmentations. Our core idea is to separate the data into two independent, but complementary views and align the representations of one view to mirror the nearest-neighbor correspondence of the data in the second view. Our experimental evaluation supported by a theoretical analysis justifies that SeBA generates an output space, which improves the feature-label relationship. An experimental study conducted in various benchmark datasets demonstrates that SeBA achieves the state-of-the-art performance in the majority of cases, opening a new avenue for SS-FSL paradigm in the domain of tabular data.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Referee Report

3 major / 2 minor

Summary. The paper proposes Separated-at-Birth Alignment (SeBA), a joint-embedding framework for semi-supervised few-shot learning (SS-FSL) on tabular data. It splits input features into two independent but complementary views and aligns the learned representation of one view to the nearest-neighbor correspondences observed in the second view, thereby eliminating reliance on data augmentations. The central claims are that a theoretical analysis justifies an improved feature-label relationship in the output space and that experiments on benchmark datasets show SeBA achieving state-of-the-art performance in the majority of cases.

Significance. If the central claims hold, the work offers a meaningful alternative to augmentation-based SSL for tabular SS-FSL, addressing a practical limitation in domains such as medicine and finance where semantic-preserving augmentations are hard to define. The attempt to supply both theoretical justification and empirical comparisons on multiple benchmarks is a positive contribution, though the absence of detailed derivations and experimental controls reduces the immediate strength of the result.

major comments (3)
  1. [§4 (Theoretical Analysis)] §4 (Theoretical Analysis): The claim that SeBA produces an output space with improved feature-label relationships is undermined by circularity; the alignment objective is defined directly in terms of nearest-neighbor structure that is itself generated from the representations being learned, so the analysis does not independently demonstrate a reduction in any explicit measure such as conditional entropy or margin violation.
  2. [§3 (Method)] §3 (Method): The procedure for partitioning tabular columns into two independent yet complementary views is not specified (random, type-based, learned, or otherwise), yet this split is load-bearing for the assumption that nearest-neighbor relations in the second view supply a reliable supervisory signal without semantic distortion; for correlated or redundant features the split can readily produce uninformative or harmful correspondences.
  3. [§5 (Experiments)] §5 (Experiments): The SOTA claim rests on benchmark results, but the manuscript supplies neither error bars, explicit train/validation/test splits, nor a complete enumeration of baselines and hyper-parameter choices, making it impossible to verify robustness or rule out post-hoc selection effects.
minor comments (2)
  1. [§3] Notation for the two views and the alignment loss should be introduced with explicit equations early in §3 to improve readability.
  2. [Abstract / §1] The abstract and introduction would benefit from a concise statement of the exact split rule once it is defined in the method section.

Simulated Author's Rebuttal

3 responses · 0 unresolved

We thank the referee for the thoughtful and constructive report. The comments identify important areas for clarification and strengthening. We address each major comment below and will revise the manuscript to incorporate the necessary changes and additional details.

read point-by-point responses

  1. Referee: [§4 (Theoretical Analysis)] §4 (Theoretical Analysis): The claim that SeBA produces an output space with improved feature-label relationships is undermined by circularity; the alignment objective is defined directly in terms of nearest-neighbor structure that is itself generated from the representations being learned, so the analysis does not independently demonstrate a reduction in any explicit measure such as conditional entropy or margin violation.

    Authors: We appreciate the referee pointing out the risk of circularity. The analysis in §4 is constructed around a cross-view setup: nearest-neighbor correspondences are computed from the second view and used to supervise alignment of the first view's representation. Because the two views are formed by disjoint feature partitions, the supervisory signal is not derived from the same representation being optimized. Nevertheless, we acknowledge that joint training introduces some dependence and that the current derivation does not explicitly bound conditional entropy or margin quantities. In the revision we will expand §4 with a clearer statement of the cross-view independence assumption, an explicit derivation relating the alignment loss to a reduction in label-conditional entropy in the joint output space, and a short discussion of the remaining limitations of the analysis. revision: partial

  2. Referee: [§3 (Method)] §3 (Method): The procedure for partitioning tabular columns into two independent yet complementary views is not specified (random, type-based, learned, or otherwise), yet this split is load-bearing for the assumption that nearest-neighbor relations in the second view supply a reliable supervisory signal without semantic distortion; for correlated or redundant features the split can readily produce uninformative or harmful correspondences.

    Authors: The referee is correct that the partitioning procedure must be stated explicitly. In the current manuscript the split is performed by randomly assigning columns to two equal-sized groups; the assumption is that tabular data typically contain redundant or correlated features so that each group remains informative. We agree that this choice requires justification and that highly correlated or redundant columns could degrade the supervisory signal. In the revised version we will (i) describe the random partitioning procedure in §3, (ii) add a short discussion of its rationale and potential failure modes, and (iii) include an ablation comparing random, type-based, and correlation-aware splits on the benchmark datasets to demonstrate that performance is not overly sensitive to the exact partitioning strategy. revision: yes

  3. Referee: [§5 (Experiments)] §5 (Experiments): The SOTA claim rests on benchmark results, but the manuscript supplies neither error bars, explicit train/validation/test splits, nor a complete enumeration of baselines and hyper-parameter choices, making it impossible to verify robustness or rule out post-hoc selection effects.

    Authors: We fully agree that the experimental section must be made reproducible and that the current presentation is insufficient to support the SOTA claim. In the revision we will add: (i) mean and standard deviation over at least five independent runs with different random seeds, (ii) explicit description of the train/validation/test splits (including any stratification or public split identifiers) for every dataset, (iii) a complete table listing all baselines together with the hyper-parameter values or search ranges used, and (iv) the hyper-parameter grid searched for SeBA itself. These additions will allow readers to assess both statistical significance and the possibility of selection effects. revision: yes

Circularity Check

0 steps flagged

No significant circularity detected; derivation relies on independent experimental and theoretical support.

full rationale

The abstract and description define SeBA via an explicit split into complementary views followed by NN-based alignment of representations, then assert improved feature-label relationship via separate experimental evaluation plus theoretical analysis. No equations, self-citations, or fitted parameters are shown reducing the central claim to the input definition by construction. Benchmarks provide external grounding, and the split/alignment procedure is presented as a modeling choice rather than a tautology. This is the normal non-circular case.

Axiom & Free-Parameter Ledger

0 free parameters · 1 axioms · 1 invented entities

The central claim rests on the domain assumption that tabular data admits a meaningful split into two independent yet complementary views and on the new alignment procedure itself; no free parameters or invented physical entities are mentioned in the abstract.

axioms (1)
  • domain assumption Tabular data can be separated into two independent but complementary views without loss of semantic information.
    This separation is the foundational premise required for the SeBA alignment to be well-defined.
invented entities (1)
  • Separated-at-Birth Alignment (SeBA) no independent evidence
    purpose: Joint-embedding framework that aligns one view's representations to the nearest-neighbor structure of the other view.
    Newly introduced method whose only evidence is the paper's own theoretical analysis and experiments.

pith-pipeline@v0.9.0 · 5536 in / 1535 out tokens · 63738 ms · 2026-05-12T01:26:29.895969+00:00 · methodology

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Lean theorems connected to this paper

Citations machine-checked in the Pith Canon. Every link opens the source theorem in the public Lean library.

What do these tags mean?
matches
The paper's claim is directly supported by a theorem in the formal canon.
supports
The theorem supports part of the paper's argument, but the paper may add assumptions or extra steps.
extends
The paper goes beyond the formal theorem; the theorem is a base layer rather than the whole result.
uses
The paper appears to rely on the theorem as machinery.
contradicts
The paper's claim conflicts with a theorem or certificate in the canon.
unclear
Pith found a possible connection, but the passage is too broad, indirect, or ambiguous to say the theorem truly supports the claim.

Reference graph

Works this paper leans on

299 extracted references · 299 canonical work pages · 6 internal anchors

  1. [1]

    Scaling Learning Algorithms Towards

    Bengio, Yoshua and LeCun, Yann , booktitle =. Scaling Learning Algorithms Towards

    work page

  2. [2]

    and Osindero, Simon and Teh, Yee Whye , journal =

    Hinton, Geoffrey E. and Osindero, Simon and Teh, Yee Whye , journal =. A Fast Learning Algorithm for Deep Belief Nets , volume =

    work page

  3. [3]

    2016 , publisher=

    Deep learning , author=. 2016 , publisher=

    work page 2016

  4. [4]

    Jingtao Li and Lingjuan Lyu and Daisuke Iso and Chaitali Chakrabarti and Michael Spranger , booktitle=. Moco. 2023 , url=

    work page 2023

  5. [5]

    Deep medicine: how artificial intelligence can make healthcare human again , author=

    work page

  6. [6]

    International Conference on Machine Learning , pages=

    Weight uncertainty in neural network , author=. International Conference on Machine Learning , pages=. 2015 , organization=

    work page 2015

  7. [7]

    IEEE Transactions on Pattern Analysis and Machine Intelligence , year=

    OneFlow: One-class flow for anomaly detection based on a minimal volume region , author=. IEEE Transactions on Pattern Analysis and Machine Intelligence , year=

    work page

  8. [8]

    2021 International Joint Conference on Neural Networks (IJCNN) , year=

    Entropic Out-of-Distribution Detection , author=. 2021 International Joint Conference on Neural Networks (IJCNN) , year=

    work page 2021

  9. [9]

    arXiv preprint arXiv:1906.09686 , year=

    Quality of uncertainty quantification for Bayesian neural network inference , author=. arXiv preprint arXiv:1906.09686 , year=

    work page arXiv 1906

  10. [10]

    International conference on machine learning , pages=

    Probabilistic backpropagation for scalable learning of bayesian neural networks , author=. International conference on machine learning , pages=. 2015 , organization=

    work page 2015

  11. [11]

    International Conference on Learning Representations , year=

    Deep Anomaly Detection with Outlier Exposure , author=. International Conference on Learning Representations , year=

    work page

  12. [12]

    Scientific reports , volume=

    Deep patient: an unsupervised representation to predict the future of patients from the electronic health records , author=. Scientific reports , volume=. 2016 , publisher=

    work page 2016

  13. [13]

    Applied Soft Computing (under review) , year=

    Revisiting multiple instance neural networks , author=. Applied Soft Computing (under review) , year=

    work page

  14. [14]

    ICLR 2021-International Conference on Learning Representations , year=

    not-MIWAE: Deep Generative Modelling with Missing not at Random Data , author=. ICLR 2021-International Conference on Learning Representations , year=

    work page 2021

  15. [15]

    2022 IEEE/CVF Winter Conference on Applications of Computer Vision (WACV) , pages=

    MisConv: Convolutional Neural Networks for Missing Data , author=. 2022 IEEE/CVF Winter Conference on Applications of Computer Vision (WACV) , pages=. 2022 , organization=

    work page 2022

  16. [16]

    Journal of Chemical Information and Modeling , volume=

    Practical applications of deep learning to impute heterogeneous drug discovery data , author=. Journal of Chemical Information and Modeling , volume=. 2020 , publisher=

    work page 2020

  17. [17]

    Advances in neural information processing systems , volume=

    A probabilistic u-net for segmentation of ambiguous images , author=. Advances in neural information processing systems , volume=

    work page

  18. [18]

    Medical Physics , volume=

    The Lung Image Database Consortium (LIDC) and Image Database Resource Initiative (IDRI): A completed reference database of lung nodules on CT scans , author=. Medical Physics , volume=

    work page

  19. [19]

    ICLR , year=

    NBDT: neural-backed decision trees , author=. ICLR , year=

    work page

  20. [20]

    JAMA , volume=

    Diagnostic assessment of deep learning algorithms for detection of lymph node metastases in women with breast cancer , author=. JAMA , volume=. 2017 , publisher=

    work page 2017

  21. [21]

    Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , pages=

    Dual-stream multiple instance learning network for whole slide image classification with self-supervised contrastive learning , author=. Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , pages=

    work page

  22. [22]

    CVPR , year=

    Dimensionality reduction by learning an invariant mapping , author=. CVPR , year=

    work page

  23. [24]

    NeurIPS , year=

    Big self-supervised models are strong semi-supervised learners , author=. NeurIPS , year=

    work page

  24. [25]

    Bootstrap Your Own Latent - A New Approach to Self-Supervised Learning , url =

    Grill, Jean-Bastien and Strub, Florian and Altch\'. Bootstrap Your Own Latent - A New Approach to Self-Supervised Learning , url =. Advances in Neural Information Processing Systems , editor =

    work page

  25. [26]

    AAAI , year=

    Loss-based attention for deep multiple instance learning , author=. AAAI , year=

    work page

  26. [27]

    Nature Biomedical Engineering , year=

    Data-efficient and weakly supervised computational pathology on whole-slide images , author=. Nature Biomedical Engineering , year=

    work page

  27. [28]

    WACV , year=

    Kernel Self-Attention for Weakly-supervised Image Classification using Deep Multiple Instance Learning , author=. WACV , year=

    work page

  28. [29]

    arXiv:1906.04881 , year=

    Multiple instance learning with graph neural networks , author=. arXiv:1906.04881 , year=

    work page arXiv 1906

  29. [30]

    arXiv:2106.00908 , year=

    TransMIL: Transformer based Correlated Multiple Instance Learning for Whole Slide Image Classication , author=. arXiv:2106.00908 , year=

    work page arXiv

  30. [31]

    ACML , year=

    Deep multi-instance learning with dynamic pooling , author=. ACML , year=

    work page

  31. [32]

    Pattern Recognition , year=

    Revisiting multiple instance neural networks , author=. Pattern Recognition , year=

    work page

  32. [33]

    AAAI , year=

    Deep MIML network , author=. AAAI , year=

    work page

  33. [34]

    NeurIPS , year=

    Support Vector Machines for Multiple-Instance Learning , author=. NeurIPS , year=

    work page

  34. [35]

    ICML , year=

    Attention-based deep multiple instance learning , author=. ICML , year=

    work page

  35. [36]

    Intravascular Imaging and Computer Assisted Stenting and Large-Scale Annotation of Biomedical Data and Expert Label Synthesis , pages=

    Radiology Objects in COntext (ROCO): a multimodal image dataset , author=. Intravascular Imaging and Computer Assisted Stenting and Large-Scale Annotation of Biomedical Data and Expert Label Synthesis , pages=. 2018 , publisher=

    work page 2018

  36. [37]

    PloS one , year =

    Deep Learning Approach to Description and Classification of Fungi Microscopic Images , author =. PloS one , year =

    work page

  37. [38]

    PloS one , year =

    Deep Learning Approach to Bacterial Colony Classification , author =. PloS one , year =

    work page

  38. [39]

    IJCNN , year =

    Deep Learning Classification of Bacteria Clones Explained by Persistence Homology , author =. IJCNN , year =

    work page

  39. [40]

    bioRxiv , year=

    Three million images and morphological profiles of cells treated with matched chemical and genetic perturbations , author=. bioRxiv , year=

    work page

  40. [41]

    Scientific data , volume=

    A radiogenomic dataset of non-small cell lung cancer , author=. Scientific data , volume=. 2018 , publisher=

    work page 2018

  41. [42]

    The Cancer Imaging Archive , year=

    Radiology data from the cancer genome atlas kidney renal clear cell carcinoma [TCGA-KIRC] collection , author=. The Cancer Imaging Archive , year=

    work page

  42. [43]

    International Conference on Medical Image Computing and Computer-Assisted Intervention , pages=

    Learning semantics-enriched representation via self-discovery, self-classification, and self-restoration , author=. International Conference on Medical Image Computing and Computer-Assisted Intervention , pages=. 2020 , organization=

    work page 2020

  43. [44]

    Advances in Neural Information Processing Systems , volume=

    Unsupervised Part Discovery from Contrastive Reconstruction , author=. Advances in Neural Information Processing Systems , volume=

    work page

  44. [45]

    One Explanation Does Not Fit All:

    Arya and others , year =. One Explanation Does Not Fit All:

    work page

  45. [46]

    Explanation and Justification in Machine Learning:

    Biran and. Explanation and Justification in Machine Learning:

    work page

  46. [47]

    Neural machine translation by jointly learning to align and translate , booktitle =

    Bahdanau and. Neural machine translation by jointly learning to align and translate , booktitle =

    work page

  47. [48]

    Selvaraju and. Grad-. 2017 , pages =. doi:10/gfkqbw , file =

    work page 2017

  48. [49]

    2017 , journal =

    Smoothgrad: Removing Noise by Adding Noise , author =. 2017 , journal =

    work page 2017

  49. [50]

    Nucleic Acids Research , volume=

    PubChem substance and compound databases , author=. Nucleic Acids Research , volume=. 2016 , publisher=

    work page 2016

  50. [51]

    Expert opinion on drug discovery , volume=

    BindingDB and ChEMBL: online compound databases for drug discovery , author=. Expert opinion on drug discovery , volume=. 2011 , publisher=

    work page 2011

  51. [52]

    Annals of Behavioral Medicine , volume=

    Compliance with cardiovascular disease prevention strategies: a review of the research , author=. Annals of Behavioral Medicine , volume=. 1997 , publisher=

    work page 1997

  52. [53]

    International Journal of Data Warehousing and Mining (IJDWM) , volume=

    Multi-label classification: An overview , author=. International Journal of Data Warehousing and Mining (IJDWM) , volume=. 2007 , publisher=

    work page 2007

  53. [54]

    Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , pages=

    Multi-label image recognition with graph convolutional networks , author=. Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , pages=

    work page

  54. [55]

    Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , pages=

    Learning a deep convnet for multi-label classification with partial labels , author=. Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , pages=

    work page

  55. [56]

    Advances in Neural Information Processing Systems , volume=

    Exploiting weakly supervised visual patterns to learn from partial annotations , author=. Advances in Neural Information Processing Systems , volume=

    work page

  56. [57]

    Proceedings of the AAAI conference on artificial intelligence , volume=

    Chexpert: A large chest radiograph dataset with uncertainty labels and expert comparison , author=. Proceedings of the AAAI conference on artificial intelligence , volume=

    work page

  57. [58]

    International Conference on Learning Representations , year=

    Large Scale GAN Training for High Fidelity Natural Image Synthesis , author=. International Conference on Learning Representations , year=

    work page

  58. [59]

    Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , pages=

    A style-based generator architecture for generative adversarial networks , author=. Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , pages=

    work page

  59. [60]

    Semi-supervised learning with deep generative models , author=. Proc. Advances in Neural Information Processing Systems (NIPS) , pages=

    work page

  60. [61]

    arXiv preprint arXiv:1812.06190 , year=

    Learning latent subspaces in variational autoencoders , author=. arXiv preprint arXiv:1812.06190 , year=

    work page arXiv

  61. [62]

    Conditional Generative Adversarial Nets

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

    work page internal anchor Pith review Pith/arXiv arXiv

  62. [63]

    IEEE Transactions on Image Processing , volume=

    Attgan: Facial attribute editing by only changing what you want , author=. IEEE Transactions on Image Processing , volume=. 2019 , publisher=

    work page 2019

  63. [64]

    arXiv e-prints , pages=

    Styleflow: Attribute-conditioned exploration of stylegan-generated images using conditional continuous normalizing flows , author=. arXiv e-prints , pages=

    work page

  64. [65]

    Engineering Applications of Artificial Intelligence , volume=

    A novel purity-based k nearest neighbors imputation method and its application in financial distress prediction , author=. Engineering Applications of Artificial Intelligence , volume=. 2019 , publisher=

    work page 2019

  65. [66]

    arXiv preprint arXiv:1705.02737 , year=

    Multiple imputation using deep denoising autoencoders , author=. arXiv preprint arXiv:1705.02737 , year=

    work page arXiv

  66. [67]

    Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition , pages=

    Generative face completion , author=. Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition , pages=

    work page

  67. [68]

    Journal of statistical software , volume=

    mice: Multivariate imputation by chained equations in R , author=. Journal of statistical software , volume=. 2011 , publisher=

    work page 2011

  68. [69]

    International Conference on Learning Representations , year=

    Why Not to Use Zero Imputation? Correcting Sparsity Bias in Training Neural Networks , author=. International Conference on Learning Representations , year=

    work page

  69. [70]

    ICML Workshop on the Art of Learning with Missing Values (Artemiss) , year=

    How to deal with missing data in supervised deep learning? , author=. ICML Workshop on the Art of Learning with Missing Values (Artemiss) , year=

    work page

  70. [71]

    Supervised learning from incomplete data via an

    Ghahramani, Zoubin and Jordan, Michael I , booktitle=. Supervised learning from incomplete data via an. 1994 , organization=

    work page 1994

  71. [72]

    1997 , publisher=

    Analysis of incomplete multivariate data , author=. 1997 , publisher=

    work page 1997

  72. [73]

    Proceedings of the International Conference on Machine Learning , pages=

    Learning from incomplete data with infinite imputations , author=. Proceedings of the International Conference on Machine Learning , pages=. 2008 , organization=

    work page 2008

  73. [74]

    Improved Training of Wasserstein GANs , url =

    Gulrajani, Ishaan and Ahmed, Faruk and Arjovsky, Martin and Dumoulin, Vincent and Courville, Aaron C , booktitle =. Improved Training of Wasserstein GANs , url =

    work page

  74. [75]

    Advances in Neural Information Processing Systems , pages=

    Feature set embedding for incomplete data , author=. Advances in Neural Information Processing Systems , pages=

    work page

  75. [76]

    Proceedings of the International Conference on Machine Learning , pages=

    Nightmare at test time: robust learning by feature deletion , author=. Proceedings of the International Conference on Machine Learning , pages=. 2006 , organization=

    work page 2006

  76. [77]

    Machine Learning , volume=

    Learning to classify with missing and corrupted features , author=. Machine Learning , volume=. 2010 , publisher=

    work page 2010

  77. [78]

    2007 , publisher=

    Missing data: A gentle introduction , author=. 2007 , publisher=

    work page 2007

  78. [79]

    Journal of chemical information and modeling , volume=

    Novel 2D fingerprints for ligand-based virtual screening , author=. Journal of chemical information and modeling , volume=. 2006 , publisher=

    work page 2006

  79. [80]

    Proceedings of the IEEE Computer Society Conference on Computer Vision and Pattern Recognition , pages=

    Shape matching and object recognition using low distortion correspondences , author=. Proceedings of the IEEE Computer Society Conference on Computer Vision and Pattern Recognition , pages=. 2005 , organization=

    work page 2005

  80. [81]

    ACM Transactions on Intelligent Systems and Technology , volume=

    LIBSVM: A Library for Support Vector Machines , author=. ACM Transactions on Intelligent Systems and Technology , volume=. 2011 , publisher=

    work page 2011

Showing first 80 references.