Pith. sign in

REVIEW 2 major objections 2 minor 72 references

Tuning only the classifier recovers a median 86% of the gains from jointly optimizing a semi-supervised security classification pipeline.

Reviewed by Pith at T0; open to challenge. T0 means a machine referee read the full paper against a public rubric. the ladder, T0–T4 →

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

T0 review · grok-4.3

2026-07-02 19:56 UTC pith:M3HLGN7W

load-bearing objection On these five security datasets, tuning the classifier alone recovers most of the joint SSL optimization benefit under equal budget, but the equivalence rests on narrow scope and an unmotivated effect-size bound. the 2 major comments →

arxiv 2607.00113 v1 pith:M3HLGN7W submitted 2026-06-30 cs.LG

SemiScope: Disentangling Classifier Tuning and Joint Optimization in Semi-Supervised Security Classification

Rui Shu , Tianpei Xia , Jingzhu He This is my paper
classification cs.LG
keywords semi-supervised learningsecurity classificationhyperparameter optimizationself-trainingclassifier tuningBayesian optimizationtabular dataequivalence testing
verification ladder T0 review T1 audit T2 compute T3 formal T4 reserved

The pith

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

The paper introduces SemiScope to jointly optimize SSL parameters, confidence filtering, oversampling, and the classifier via Bayesian optimization. It compares this to Tuned-Clf, which applies the same classifier tuning budget but leaves SSL at default settings. On five tabular security datasets with 10% labeled data, Tuned-Clf proves statistically equivalent to the full pipeline on four datasets using paired equivalence tests with a one-point g-measure threshold. This shows that most reported benefits of complex SSL tuning stem from downstream classifier optimization rather than interactions with the semi-supervised components.

Core claim

Classifier hyperparameter optimization alone recovers a median 86% of SemiScope's gain over default self-training plus random forest. Under an equal-budget control, the tuned classifier alone reaches statistical equivalence with the full joint-optimization pipeline on four of the five datasets examined.

What carries the argument

The Tuned-Clf control, which fixes SSL components to defaults while allocating the full Bayesian optimization budget to the classifier and validation threshold tuning.

Load-bearing premise

The five tabular security datasets together with the g-measure and +/-1.0 equivalence threshold adequately represent the broader space of security classification problems.

What would settle it

Running the same protocol on a sixth security dataset and finding that Tuned-Clf falls short of SemiScope by more than one g-measure point.

Watch this falsifier — get emailed when new claim-graph text bears on it.

If this is right

  • Self-training combined with classifier tuning and threshold tuning reaches within one g-measure of fully supervised random forest at 20-30% labels on four datasets.
  • The simpler recipe performs at or better than default self-training plus random forest at the same or lower label rates.
  • The decomposition protocol isolates the contribution of each pipeline component and can be applied to other SSL settings.
  • Joint search over SSL and classifier yields only marginal additional gains beyond classifier tuning in this domain.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

  • Future work on SSL for security data could prioritize classifier families and their tuning over developing new pseudo-labeling schemes.
  • The finding raises the question whether similar patterns hold when using neural networks or graph-based SSL methods instead of tree classifiers.
  • Budget allocation in security ML pipelines should favor classifier search when label scarcity is the main constraint.
  • Replicating the protocol on additional datasets would test whether the 86% recovery rate generalizes beyond the five chosen security tasks.

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

2 major / 2 minor

Summary. The manuscript introduces SemiScope as an analysis instrument that uses Bayesian optimization to jointly tune SSL components (self-training defaults, confidence filtering, oversampling) together with classifier hyperparameters and decision threshold for binary tabular security classification. On five datasets at 10% label rate, it reports that SemiScope outperforms default SSL baselines by 0.7-12.7 g-measure points, yet a Tuned-Clf control (SSL fixed to defaults but given identical 100-trial BO budget and validation threshold tuning) is statistically equivalent on 4/5 datasets under paired TOST with smallest effect of interest +/-1.0 g-measure and recovers a median 86% of SemiScope's gain over Default ST+RF. The authors conclude that the reusable contribution is the decomposition protocol and that a simpler recipe (Self-Training + classifier BO + threshold tuning) suffices.

Significance. If the equivalence result holds under the stated controls, the work supplies a concrete, falsifiable protocol for attributing gains in SSL pipelines to classifier HPO versus joint interactions, which is a useful methodological contribution for label-scarce domains. The equal-budget control and explicit use of TOST equivalence testing (rather than superiority tests) are clear strengths that allow direct comparison of the two regimes.

major comments (2)
  1. [Results] Results section: The central claim that 'classifier HPO alone recovers a median 86% of SemiScope's gain' and that Tuned-Clf is 'statistically equivalent ... on 4 of 5 datasets' is obtained via paired TOST with a smallest effect of interest fixed at +/-1.0 g-measure. The manuscript supplies no justification, domain-specific rationale, or sensitivity analysis for this particular delta; altering it would directly change whether equivalence is declared and therefore whether the 'simpler recipe suffices' conclusion follows.
  2. [Results] Results / Conclusions: The recommendation that the simpler Self-Training + tuned-classifier recipe 'reaches within 1 g-measure of Supervised RF at 20-30% labels on four datasets' is extrapolated from the five chosen tabular security datasets. No selection criteria, diversity argument, or discussion of how these datasets represent the broader class of security classification problems is provided, which is load-bearing for the scope of the general claim.
minor comments (2)
  1. [Abstract] Abstract: The range '0.7-12.7 points' is reported without stating the metric (g-measure) or identifying the strongest baseline in the sentence, which reduces immediate readability.
  2. The paper does not report the exact feature preprocessing steps or how the 100-trial budget interacts with the inner validation-set threshold tuning loop, which would aid reproducibility of the equal-budget control.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for the constructive review and for highlighting the methodological strengths of the equal-budget control and TOST equivalence testing. We address each major comment below and will revise the manuscript accordingly.

read point-by-point responses

  1. Referee: [Results] Results section: The central claim that 'classifier HPO alone recovers a median 86% of SemiScope's gain' and that Tuned-Clf is 'statistically equivalent ... on 4 of 5 datasets' is obtained via paired TOST with a smallest effect of interest fixed at +/-1.0 g-measure. The manuscript supplies no justification, domain-specific rationale, or sensitivity analysis for this particular delta; altering it would directly change whether equivalence is declared and therefore whether the 'simpler recipe suffices' conclusion follows.

    Authors: We agree that the manuscript provides no explicit justification or sensitivity analysis for the smallest effect of interest (SEOI) of +/-1.0 g-measure. This value was selected because a 1-point change in g-measure corresponds to a practically noticeable shift in detection performance for the security tasks considered (e.g., changes in true-positive/false-positive trade-offs that affect operational alerting). Nevertheless, the absence of supporting rationale and sensitivity checks is a valid concern. In revision we will (i) state the domain-motivated rationale for SEOI = 1.0, (ii) add a sensitivity table showing equivalence declarations for SEOI values of 0.5, 1.0, and 2.0, and (iii) qualify the 'simpler recipe suffices' claim to note its dependence on the chosen SEOI. These additions will appear in the Results section and a new appendix. revision: yes

  2. Referee: [Results] Results / Conclusions: The recommendation that the simpler Self-Training + tuned-classifier recipe 'reaches within 1 g-measure of Supervised RF at 20-30% labels on four datasets' is extrapolated from the five chosen tabular security datasets. No selection criteria, diversity argument, or discussion of how these datasets represent the broader class of security classification problems is provided, which is load-bearing for the scope of the general claim.

    Authors: We concur that the manuscript does not articulate dataset selection criteria or discuss representativeness. The five datasets were chosen because they are established public benchmarks in the tabular security-ML literature (covering phishing, Android malware, network intrusion, and spam), exhibit varying sizes, class imbalance ratios, and feature types typical of security tabular data, and have been used in prior SSL security studies. In revision we will expand the Datasets subsection with explicit selection criteria, a short diversity argument (size, imbalance, threat type), and an explicit limitations paragraph noting that results may not generalize to non-tabular or non-security domains. The scope of the 'simpler recipe' claim will be qualified accordingly. revision: yes

Circularity Check

0 steps flagged

No circularity in empirical comparison of optimization regimes

full rationale

The paper conducts a controlled empirical study that directly compares the full SemiScope joint optimization pipeline against an independent Tuned-Clf control (same classifier budget, default SSL, validation threshold tuning) using paired TOST equivalence tests on five fixed datasets. The reported median 86% recovery and equivalence on 4/5 datasets are outcomes of these separate runs, not quantities defined in terms of each other by construction, fitted parameters renamed as predictions, or load-bearing self-citations. No equations, ansatzes, or uniqueness theorems are invoked that reduce the central claims to the inputs.

Axiom & Free-Parameter Ledger

2 free parameters · 2 axioms · 0 invented entities

The central empirical claim rests on standard statistical assumptions for equivalence testing and on the representativeness of the five security datasets; no new entities are postulated and the tuned hyperparameters are treated as experimental controls rather than free parameters that define the result.

free parameters (2)
  • Bayesian optimization trial budget = 100
    The 100-trial limit is chosen by the authors and directly affects how much performance the Tuned-Clf control can recover.
  • TOST smallest effect of interest = 1.0
    The +/-1.0 g-measure threshold is selected to declare equivalence.
axioms (2)
  • domain assumption g-measure is the appropriate primary metric for imbalanced binary security classification
    All comparisons and equivalence tests are performed with g-measure.
  • domain assumption The five tabular security datasets are representative of the target application class
    Results are generalized from these datasets to security classification broadly.

pith-pipeline@v0.9.1-grok · 5892 in / 1459 out tokens · 31174 ms · 2026-07-02T19:56:34.935276+00:00 · methodology

0 comments
read the original abstract

Background. Labeled data for security classification is scarce. Semi-supervised learning (SSL) propagates labels from a small labeled pool to larger unlabeled pools. Yet security applications often use SSL as a black box: default parameters, a fixed classifier, and no handling of pseudo-label-induced class imbalance. Aims. Recent work reports sizeable gains from optimizing SSL pipelines via joint search, AutoML, or per-component tuning. These gains are hard to attribute: they may reflect useful SSL-classifier interactions, or mostly from simply tuning the downstream classifier. We disentangle these effects for binary tabular security data with classical SSL and tree-based classifiers. Method. We build SemiScope as an analysis instrument, not a deployment recommendation. It uses Bayesian Optimization to jointly tune SSL settings, confidence filtering, oversampling, and the classifier. The key control, Tuned-Clf, fixes SSL to defaults but gets the same 100-trial classifier budget and validation-set threshold tuning as SemiScope. At 10% labels, we compare them with paired TOST using a +/-1.0 g-measure smallest effect of interest. Results. SemiScope beats every default SSL baseline on all five datasets, improving over the strongest by 0.7-12.7 points. Under the equal-budget control, Tuned-Clf is statistically equivalent to the full pipeline on 4 of 5 datasets; Phishing is inconclusive. Classifier HPO alone recovers a median 86% of SemiScope's gain over Default Self-Training (ST) + Random Forest (RF). Conclusions. The reusable contribution is the decomposition protocol. A simpler recipe suffices: use Self-Training, tune the classifier with Bayesian Optimization, and tune the decision threshold on validation data. It reaches within 1 g-measure of Supervised RF at 20-30% labels on four datasets and 40% on Drebin, at the same or lower label rate than Default ST + RF on every dataset.

discussion (0)

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

Reference graph

Works this paper leans on

72 extracted references · 72 canonical work pages · 1 internal anchor

  1. [1]

    better data

    Is" better data" better than" better data miners"? on the benefits of tuning SMOTE for defect prediction , author=. Proceedings of the 40th International Conference on Software engineering , pages=

    work page

  2. [2]

    Proceedings of the 25th ACM International on Conference on Information and Knowledge Management , pages=

    Content-agnostic malware detection in heterogeneous malicious distribution graph , author=. Proceedings of the 25th ACM International on Conference on Information and Knowledge Management , pages=

    work page

  3. [3]

    Advances in neural information processing systems , volume=

    Algorithms for hyper-parameter optimization , author=. Advances in neural information processing systems , volume=

    work page

  4. [4]

    , author=

    Random search for hyper-parameter optimization. , author=. Journal of machine learning research , volume=

    work page

  5. [5]

    , author=

    Hyperopt: A python library for optimizing the hyperparameters of machine learning algorithms. , author=. SciPy , volume=

    work page

  6. [6]

    Proceedings of the eleventh annual conference on Computational learning theory , pages=

    Combining labeled and unlabeled data with co-training , author=. Proceedings of the eleventh annual conference on Computational learning theory , pages=

    work page

  7. [7]

    2016 , publisher=

    Unsupervised learning algorithms , author=. 2016 , publisher=

    work page 2016

  8. [8]

    Journal of artificial intelligence research , volume=

    SMOTE: synthetic minority over-sampling technique , author=. Journal of artificial intelligence research , volume=

    work page

  9. [9]

    International conference on intelligent computing , pages=

    Borderline-SMOTE: a new over-sampling method in imbalanced data sets learning , author=. International conference on intelligent computing , pages=. 2005 , organization=

    work page 2005

  10. [10]

    2015 IEEE international conference on communications (ICC) , pages=

    6 million spam tweets: A large ground truth for timely Twitter spam detection , author=. 2015 IEEE international conference on communications (ICC) , pages=. 2015 , organization=

    work page 2015

  11. [11]

    Proceedings of the 22nd acm sigkdd international conference on knowledge discovery and data mining , pages=

    Xgboost: A scalable tree boosting system , author=. Proceedings of the 22nd acm sigkdd international conference on knowledge discovery and data mining , pages=

    work page

  12. [12]

    Journal of Big Data , volume=

    Survey of review spam detection using machine learning techniques , author=. Journal of Big Data , volume=. 2015 , publisher=

    work page 2015

  13. [13]

    2016 International conference on information science and security (ICISS) , pages=

    An evaluation framework for intrusion detection dataset , author=. 2016 International conference on information science and security (ICISS) , pages=. 2016 , organization=

    work page 2016

  14. [14]

    2009 , publisher=

    The elements of statistical learning: data mining, inference, and prediction , author=. 2009 , publisher=

    work page 2009

  15. [15]

    International conference on learning and intelligent optimization , pages=

    Sequential model-based optimization for general algorithm configuration , author=. International conference on learning and intelligent optimization , pages=. 2011 , organization=

    work page 2011

  16. [16]

    2021 , howpublished=

    work page 2021

  17. [17]

    Proceedings of the IEEE/CVF conference on computer vision and pattern recognition , pages=

    Label propagation for deep semi-supervised learning , author=. Proceedings of the IEEE/CVF conference on computer vision and pattern recognition , pages=

    work page

  18. [18]

    International Conference on Detection of Intrusions and Malware, and Vulnerability Assessment , pages=

    Adaptive semantics-aware malware classification , author=. International Conference on Detection of Intrusions and Malware, and Vulnerability Assessment , pages=. 2016 , organization=

    work page 2016

  19. [19]

    2019 IFIP/IEEE Symposium on Integrated Network and Service Management (IM) , pages=

    Machine learning based insider threat modelling and detection , author=. 2019 IFIP/IEEE Symposium on Integrated Network and Service Management (IM) , pages=. 2019 , organization=

    work page 2019

  20. [20]

    Workshop on challenges in representation learning, ICML , volume=

    Pseudo-label: The simple and efficient semi-supervised learning method for deep neural networks , author=. Workshop on challenges in representation learning, ICML , volume=. 2013 , organization=

    work page 2013

  21. [21]

    , author=

    Optimizing Hyperparameters of Support Vector Machines by Genetic Algorithms. , author=. IC-AI , volume=

    work page

  22. [22]

    Proceedings of the genetic and evolutionary computation conference , pages=

    Particle swarm optimization for hyper-parameter selection in deep neural networks , author=. Proceedings of the genetic and evolutionary computation conference , pages=

    work page

  23. [23]

    International conference on network and system security , pages=

    Detecting malicious urls using lexical analysis , author=. International conference on network and system security , pages=. 2016 , organization=

    work page 2016

  24. [24]

    2016 IEEE Conference on Intelligence and Security Informatics (ISI) , pages=

    Darknet and deepnet mining for proactive cybersecurity threat intelligence , author=. 2016 IEEE Conference on Intelligence and Security Informatics (ISI) , pages=. 2016 , organization=

    work page 2016

  25. [25]

    2016 IEEE International Conference on Big Data (Big Data) , pages=

    Label propagation in big data to detect remote access Trojans , author=. 2016 IEEE International Conference on Big Data (Big Data) , pages=. 2016 , organization=

    work page 2016

  26. [26]

    ACM Computing Surveys (CSUR) , volume=

    A survey of random forest based methods for intrusion detection systems , author=. ACM Computing Surveys (CSUR) , volume=. 2018 , publisher=

    work page 2018

  27. [27]

    the Journal of machine Learning research , volume=

    Scikit-learn: Machine learning in Python , author=. the Journal of machine Learning research , volume=. 2011 , publisher=

    work page 2011

  28. [28]

    IEEE Transactions on Information Theory , volume=

    Probability of error of some adaptive pattern-recognition machines , author=. IEEE Transactions on Information Theory , volume=. 1965 , publisher=

    work page 1965

  29. [29]

    proceedings of the 2008 conference on empirical methods in natural language processing , pages=

    An analysis of active learning strategies for sequence labeling tasks , author=. proceedings of the 2008 conference on empirical methods in natural language processing , pages=

    work page 2008

  30. [30]

    Proceedings of the IEEE , volume=

    Taking the human out of the loop: A review of Bayesian optimization , author=. Proceedings of the IEEE , volume=. 2015 , publisher=

    work page 2015

  31. [31]

    , author=

    Toward generating a new intrusion detection dataset and intrusion traffic characterization. , author=. ICISSp , volume=

    work page

  32. [32]

    ACM Computing Surveys (CSUR) , volume=

    A survey and comparative study of tweet sentiment analysis via semi-supervised learning , author=. ACM Computing Surveys (CSUR) , volume=. 2016 , publisher=

    work page 2016

  33. [33]

    Advances in neural information processing systems , volume=

    Practical bayesian optimization of machine learning algorithms , author=. Advances in neural information processing systems , volume=

    work page

  34. [34]

    Human-centric Computing and Information Sciences , volume=

    A state-of-the-art survey of malware detection approaches using data mining techniques , author=. Human-centric Computing and Information Sciences , volume=. 2018 , publisher=

    work page 2018

  35. [35]

    Journal of global optimization , volume=

    Differential evolution--a simple and efficient heuristic for global optimization over continuous spaces , author=. Journal of global optimization , volume=. 1997 , publisher=

    work page 1997

  36. [36]

    Pacific-Asia conference on knowledge discovery and data mining , pages=

    Defending against backdoor attacks by layer-wise feature analysis , author=. Pacific-Asia conference on knowledge discovery and data mining , pages=. 2023 , organization=

    work page 2023

  37. [37]

    IEEE Transactions on Software Engineering , volume=

    Better data labelling with emblem (and how that impacts defect prediction) , author=. IEEE Transactions on Software Engineering , volume=. 2020 , publisher=

    work page 2020

  38. [38]

    Machine learning , volume=

    A survey on semi-supervised learning , author=. Machine learning , volume=. 2020 , publisher=

    work page 2020

  39. [39]

    24th USENIX Security Symposium (USENIX Security 15) , pages=

    \ EASEAndroid \ : Automatic policy analysis and refinement for security enhanced android via \ Large-Scale \ \ Semi-Supervised \ learning , author=. 24th USENIX Security Symposium (USENIX Security 15) , pages=

    work page

  40. [40]

    Proceedings of the Twenty-Ninth International Conference on International Joint Conferences on Artificial Intelligence , pages=

    Collaboration based multi-label propagation for fraud detection , author=. Proceedings of the Twenty-Ninth International Conference on International Joint Conferences on Artificial Intelligence , pages=

    work page

  41. [41]

    Computers & Security , volume=

    Twitter spam detection: Survey of new approaches and comparative study , author=. Computers & Security , volume=. 2018 , publisher=

    work page 2018

  42. [42]

    Neurocomputing , volume=

    On hyperparameter optimization of machine learning algorithms: Theory and practice , author=. Neurocomputing , volume=. 2020 , publisher=

    work page 2020

  43. [43]

    Proceedings of the First Workshop on Insights from Negative Results in NLP , pages=

    Label propagation-based semi-supervised learning for hate speech classification , author=. Proceedings of the First Workshop on Insights from Negative Results in NLP , pages=

    work page

  44. [44]

    Automated Software Engineering , volume=

    Label propagation based semi-supervised learning for software defect prediction , author=. Automated Software Engineering , volume=. 2017 , publisher=

    work page 2017

  45. [45]

    Advances in neural information processing systems , volume=

    Learning with local and global consistency , author=. Advances in neural information processing systems , volume=

    work page

  46. [46]

    ProQuest number: information to all users , year=

    Learning from labeled and unlabeled data with label propagation , author=. ProQuest number: information to all users , year=

    work page

  47. [47]

    2009 , publisher=

    Introduction to semi-supervised learning , author=. 2009 , publisher=

    work page 2009

  48. [48]

    2005 , publisher=

    Semi-supervised learning literature survey , author=. 2005 , publisher=

    work page 2005

  49. [49]

    Proceedings of the 25th ACM SIGKDD international conference on knowledge discovery & data mining , pages=

    Optuna: A next-generation hyperparameter optimization framework , author=. Proceedings of the 25th ACM SIGKDD international conference on knowledge discovery & data mining , pages=

    work page

  50. [50]

    , author=

    Drebin: Effective and explainable detection of android malware in your pocket. , author=. Ndss , volume=. 2014 , organization=

    work page 2014

  51. [51]

    Machine learning , volume=

    Random forests , author=. Machine learning , volume=. 2001 , publisher=

    work page 2001

  52. [52]

    Advances in neural information processing systems , volume=

    Efficient and robust automated machine learning , author=. Advances in neural information processing systems , volume=

    work page

  53. [53]

    Advances in neural information processing systems , volume=

    Lightgbm: A highly efficient gradient boosting decision tree , author=. Advances in neural information processing systems , volume=

    work page

  54. [54]

    2008 3rd international conference on information and communication technologies: from theory to applications , pages=

    Intelligent phishing website detection system using fuzzy techniques , author=. 2008 3rd international conference on information and communication technologies: from theory to applications , pages=. 2008 , organization=

    work page 2008

  55. [55]

    2015 military communications and information systems conference (MilCIS) , pages=

    UNSW-NB15: a comprehensive data set for network intrusion detection systems (UNSW-NB15 network data set) , author=. 2015 military communications and information systems conference (MilCIS) , pages=. 2015 , organization=

    work page 2015

  56. [56]

    Workshop on automatic machine learning , pages=

    TPOT: A tree-based pipeline optimization tool for automating machine learning , author=. Workshop on automatic machine learning , pages=. 2016 , organization=

    work page 2016

  57. [57]

    arXiv preprint arXiv:2205.00665 , year=

    Reducing the Cost of Training Security Classifier (via Optimized Semi-Supervised Learning) , author=. arXiv preprint arXiv:2205.00665 , year=

    work page arXiv

  58. [58]

    Advances in neural information processing systems , volume=

    Fixmatch: Simplifying semi-supervised learning with consistency and confidence , author=. Advances in neural information processing systems , volume=

    work page

  59. [59]

    2009 IEEE symposium on computational intelligence for security and defense applications , pages=

    A detailed analysis of the KDD CUP 99 data set , author=. 2009 IEEE symposium on computational intelligence for security and defense applications , pages=. 2009 , organization=

    work page 2009

  60. [60]

    Proceedings of machine learning and systems , volume=

    Flaml: A fast and lightweight automl library , author=. Proceedings of machine learning and systems , volume=

    work page

  61. [61]

    Advances in neural information processing systems , volume=

    Flexmatch: Boosting semi-supervised learning with curriculum pseudo labeling , author=. Advances in neural information processing systems , volume=

    work page

  62. [62]

    IEEE Transactions on Software Engineering , volume=

    Finding faster configurations using flash , author=. IEEE Transactions on Software Engineering , volume=. 2018 , publisher=

    work page 2018

  63. [63]

    2017 32nd IEEE/ACM International Conference on Automated Software Engineering (ASE) , pages=

    Transfer learning for performance modeling of configurable systems: An exploratory analysis , author=. 2017 32nd IEEE/ACM International Conference on Automated Software Engineering (ASE) , pages=. 2017 , organization=

    work page 2017

  64. [64]

    ACM Computing Surveys , volume=

    A Survey on the Applications of Semi-supervised Learning to Cyber-security , author=. ACM Computing Surveys , volume=. 2024 , publisher=

    work page 2024

  65. [65]

    31st USENIX Security Symposium (USENIX Security 22) , pages=

    Dos and don'ts of machine learning in computer security , author=. 31st USENIX Security Symposium (USENIX Security 22) , pages=

    work page

  66. [66]

    arXiv preprint arXiv:2205.07246 (2022)

    Freematch: Self-adaptive thresholding for semi-supervised learning , author=. arXiv preprint arXiv:2205.07246 , year=

    work page arXiv

  67. [67]

    Advances in neural information processing systems , volume=

    Vime: Extending the success of self-and semi-supervised learning to tabular domain , author=. Advances in neural information processing systems , volume=

    work page

  68. [68]

    Advances in Neural Information Processing Systems , volume=

    Subtab: Subsetting features of tabular data for self-supervised representation learning , author=. Advances in Neural Information Processing Systems , volume=

    work page

  69. [69]

    Social psychological and personality science , volume=

    Equivalence tests: A practical primer for t tests, correlations, and meta-analyses , author=. Social psychological and personality science , volume=. 2017 , publisher=

    work page 2017

  70. [70]

    SCARF: Self-Supervised Contrastive Learning using Random Feature Corruption

    Scarf: Self-supervised contrastive learning using random feature corruption , author=. arXiv preprint arXiv:2106.15147 , year=

    work page Pith review arXiv

  71. [71]

    TabPFN: A Transformer That Solves Small Tabular Classification Problems in a Second

    Tabpfn: A transformer that solves small tabular classification problems in a second , author=. arXiv preprint arXiv:2207.01848 , year=

    work page internal anchor Pith review Pith/arXiv arXiv

  72. [72]

    28th USENIX security symposium (USENIX Security 19) , pages=

    \ TESSERACT \ : Eliminating experimental bias in malware classification across space and time , author=. 28th USENIX security symposium (USENIX Security 19) , pages=

    work page