Pith. sign in

REVIEW 4 major objections 5 minor 41 references

GradStop: Exploring Training Dynamics in Unsupervised Outlier Detection through Gradient

T0 review · 4 major / 5 minor · reviewed 2026-08-11 · deepseek-v4-flash

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

Pith's one-line read A label-free training-dynamics method, GradStop, aims to stop deep unsupervised outlier detection models at the epoch where detection performance peaks, preventing the degradation that comes from overfitting to outliers.

desk verdict The AE improvement is real but mostly comes from a stop-at-init divergence rule, not the advertised gradient-cohesion monitoring; the paper overstates its central mechanism, yet the underlying idea and evaluation are worth a serious referee. read the letter →

arxiv 2412.08501 v2 pith:BGF4GJMB submitted 2024-12-11 cs.LG

Yuang Zhang , Liping Wang , Yihong Huang , Yuanxing Zheng , Fan Zhang , Xuemin Lin This is my paper · ORCID
classification cs.LG
keywords unsupervisedoutlierdetectionearlystoppingtrainingdynamicsgradientcohesioninlierpriorityautoencoderDeepSVDDsampling
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

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

The reading

The paper claims that a deep unsupervised outlier detection model's training can be stopped at the right epoch without any labels by watching the geometry of per-sample gradients. During training, inliers fit faster and their gradients stay more aligned than outliers' gradients, a phenomenon called inlier priority; when that alignment gap closes, further training hurts detection. GradStop samples the k largest- and k smallest-gradient-norm points each epoch, measures their inner cohesion and inter-divergence, and halts when the cohesion gap has stopped rising for a window of epochs. Experiments on four deep UOD models and 47 real-world datasets show the rule recovers much of the performance lost to overfitting, raising AE AUC from 0.758 to 0.775 and DeepSVDD AUC from 0.502 to 0.648.

What carries the argument

The machinery has three parts: GradSample, which each epoch computes per-sample gradient norms on a fixed evaluation batch and returns the top-$k$ and bottom-$k$ gradient vectors; the cohesion metric $C(G) = \|\sum_i g_i\| \,/\, \sum_i \|g_i\|$, which measures how aligned the sampled gradients are; and the divergence metric $D(G_1, G_2)$, the angle between the two summed gradient vectors. GradStop halts training when $C(G_{\text{last}}) - C(G_{\text{top}})$ is below a benefit threshold and has not increased for $w$ epochs, or when early divergence is so large that the randomly initialized model is already better than anything training will produce.

What would settle it

On a dataset where per-sample gradient norms of inliers and outliers are statistically indistinguishable throughout training, run GradStop against random checkpoint selection; if GradStop's chosen epoch does not beat a random epoch in AUC, the ranking assumption is falsified, as the paper itself observes on datasets like SpamBase and speech where the method stops early and detection stays near chance.

Watch

Extended reading notes

Core claim

GradStop establishes that the satisfaction degree of the inlier-priority outlier assumption can be read from gradient statistics during training, and that stopping training when those statistics signal that inlier priority no longer holds mitigates the performance degradation caused by fitting outliers. The central observed quantity is the cohesion gap $C(G_{\text{last}}) - C(G_{\text{top}})$, where $G_{\text{top}}$ and $G_{\text{last}}$ are the gradient vectors of samples with the largest and smallest gradient norms; a sustained small or negative gap is taken as evidence that learning is no longer improving OD performance. The paper also provides a theoretical sufficient condition, $r_t > \cos\theta_t R + \sqrt{\cos^2\theta_t R^2 + 2R + 1}$, under which inlier priority strengthens, connecting the ratio of inlier-to-outlier gradient norms and the angle between their summed gradients to the loss-decreasing speed gap.

Load-bearing premise

The load-bearing premise is that ranking samples by per-gradient norm separates outliers from inliers well enough that the top-k set is more likely to contain outliers and the bottom-k set is more likely to contain inliers, so the cohesion gap tracks inlier priority.

Editorial extensions

If this is right

  • GradAE raises AE AUC from 0.758 to 0.775 and moves its average rank from 5.809 to 5.021 across the ten compared detection methods.
  • DeepSVDD with GradStop improves AUC from 0.502 to 0.648, a 29.08% gain, and improves detection on 38 of 47 datasets.
  • GradStop outperforms the prior label-free early stopping baseline, EntropyStop, on all four deep UOD models tested.
  • Models with constrained latent spaces, VAE and RDP, gain little from early stopping because their regularization already limits performance degradation.
  • The method needs no validation labels and runs on the contaminated training set itself, making it applicable to the common unsupervised setting.

Reading between the lines

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

  • If the cohesion gap tracks inlier priority, it could be used not just as a stopping signal but as a training signal, e.g., an auxiliary loss that encourages the gap to stay large, potentially improving the final detector beyond what early stopping alone achieves.
  • The gradient-norm ranking might serve as a source of weak labels, transforming the unsupervised problem into a self-supervised one where the top-k and bottom-k sets act as pseudo-outliers and pseudo-inliers, an idea the paper itself mentions as future work.
  • The method's dependence on the outlier-assumption's satisfaction means it may fail on datasets where the model never learns to distinguish outliers from inliers, which the paper confirms on SpamBase, speech, and WPBC; a practical safeguard would be to run GradStop in parallel with a simple baseline like Isolation Forest and compare their scores.
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 GradStop, a label-free early stopping method for deep unsupervised outlier detection (UOD). GradStop first uses GradSample to select, at each epoch, the top-k and bottom-k gradient-norm samples from an evaluation batch, treating these as collective proxies for outliers and inliers. It then computes a cohesion metric C(G) for each set and a divergence metric D between the two sums, and stops training when the cohesion difference C(Glast)-C(Gtop) is no longer increasing (indicating that inlier priority is weakening) or, at initialization, when the divergence exceeds a threshold tD, in which case the initial random model is kept. The authors report experiments on four deep UOD algorithms (AE, DeepSVDD, RDP, VAE) across 47 real-world datasets, claim improvements over vanilla training and over EntropyStop, and provide a theoretical analysis in Appendix B that derives a sufficient condition for the inlier-priority loss-decreasing gap based on gradient-norm ratios and gradient angles.

Significance. If the central claim were fully supported, the paper would make a useful practical contribution: a label-free early stopping rule for deep UOD on contaminated data, with a broad benchmark evaluation across 47 datasets and four algorithms, and publicly available code. The theoretical analysis, if it genuinely linked the implemented cohesion difference to the derived sufficient condition, would also provide an interpretable training-dynamics foundation. However, the ablation study in Appendix C.3 shows that the cohesion-based per-epoch mechanism alone improves AE from 0.758 to only 0.760, whereas the full GradAE improvement to 0.775 comes almost entirely from the divergence stop-at-init rule tD. This substantially weakens the advertised connection between the training-dynamics monitor and the reported gains, and it means the paper's main conceptual contribution is not supported by the experimental evidence as presented.

major comments (4)
  1. [Appendix C.3, Table 6] The ablation in Table 6 directly contradicts the paper's central claim that the cohesion-based training-dynamics signal drives the reported performance gains. Setting tD to infinity, which disables the divergence stop-at-init rule, leaves GradAE at 0.760 AUC versus VanillaAE at 0.758, a gain of only 0.002. In contrast, the default setting with tD=1.57 gives 0.775, an improvement of 0.017 over VanillaAE. Thus roughly 88% of the reported GradAE improvement is attributable to the one-time tD check that keeps the randomly initialized model, not to the per-epoch cohesion monitoring C(Glast)-C(Gtop) that the paper frames as the core mechanism. The authors need to either re-analyze the contribution of each component, provide evidence that the cohesion mechanism is not merely noise, or substantially reframe the paper's contribution around the divergence-based stop-at-init rule. As it stands, the central claim of Section 4.4 and the abstract is not supported by the paper's own ablation.
  2. [Appendix B, final paragraph and Theorem 4.1] The theoretical bridge between Theorem 4.1 and the implemented stopping signal is asserted rather than proved. The theorem gives a sufficient condition on rt, the ratio of the norms of summed inlier and outlier gradients, and on cos(theta_t). The final paragraph of Appendix B states that log(rt) = C(Glast)-C(Gtop) + R_tilde, where R_tilde is 'linear with R during training', and then says the stopping indicator 'is used to approximate' C(∇f_i)-C(∇f_o). No derivation or error bound is provided for this approximation, and the relation between the summed-gradient ratio rt and the average-cohesion difference C(Glast)-C(Gtop) is not established. Since this approximation is load-bearing for the claim that the algorithm implements the theory, the authors should either prove the approximation under the assumptions of Theorem 4.1, provide numerical evidence that it holds across the benchmark datasets, or explicitly state that the theory and the algorithm are connected only heuristically.
  3. [Section 5.2, Table 1 and Appendix C.2/C.3] The evaluation protocol creates a risk of overfitting the method to the same 47 datasets used for both hyperparameter selection and final reporting. The grid search in Appendix C.3 is performed over the same benchmark, and the final results are reported on those same datasets, with no held-out validation or per-dataset confidence intervals for the main aggregate numbers. Given that the tD threshold is a key driver of the reported gains, the authors should report which datasets trigger the tD stop, how the tD value was selected, and ideally evaluate the method on a separate test set of datasets or with a nested validation scheme. Otherwise, the aggregate improvement may reflect tuning to the benchmark rather than a generally applicable early stopping rule.
  4. [Appendix C.5, Tables 7 and 8] The paper's robustness claim is contradicted by several negative results that the authors acknowledge only in passing. On AE, GradStop degrades performance on fault (0.647 to 0.550, -15.04%) and ALOI (-4.99%). On DeepSVDD, GradStop degrades performance on fault, campaign, Ionosphere, shuttle, Waveform, thyroid, and Stamps, with changes as large as -29.52% on Stamps. While no method is expected to help every dataset, these cases show that the early stopping rule can stop at a substantially wrong epoch when the GradSample assumption fails, and the paper's framing of GradStop as 'effective, robust, and label-free' is too strong. The authors should provide a more systematic analysis of when the tD rule and the cohesion rule fail, including the relationship between failure and properties such as outlier proportion or the validity of the inlier-priority assumption, rather than treating these as isolated exceptions.
minor comments (5)
  1. [Appendix C.4, Figure 4 caption] The caption says 'For wine in Fig. 4b' and then 'For wine in Fig. 4c', but the third case study is on optdigits according to the surrounding text; the caption should refer to optdigits in Fig. 4c.
  2. [Algorithm 2, line 11-12] The pseudocode has formatting issues: the line 'C∆max ← max(C∆[t - w + 1 : t]' is missing a closing bracket, and the condition in line 12 mixes 'H −1' with 'C∆[t] − C∆max' in a way that is hard to parse. Please rewrite the condition with clear parentheses and consistent notation.
  3. [Appendix B, equation (14)] The derivation introduces a condition involving θt < cos^{-1}(-1/r) in equation (14) that is not used later in the proof of the main sufficient condition. Clarify whether this condition is necessary for the subsequent steps or can be omitted.
  4. [Section 5.2.1, text near Table 1] The sentence 'EntropyAE with EntropyStopHuang et al. (2024)' is missing a space between 'EntropyStop' and the citation; also, the main text refers to 'EntropyAE' while the table lists 'EntropyAE' and the related work discusses 'EntropyStop'. Please standardize the naming.
  5. [Appendix C.2] The experiments use full-batch gradient descent and a fixed evaluation batch of size 400, with GradSample recomputed every ten epochs. This is a narrow setting; a brief discussion of how the method would behave with minibatch training and with more frequent or less frequent sampling would help practitioners apply the method beyond the reported configuration.

Circularity Check

1 steps flagged · score 3.0 of 10

Moderate circularity: the headline GradAE gain is obtained after selecting GradStop's hyperparameters on the same 47 datasets used to report the improvement, and the paper's own ablation shows the cohesion mechanism contributes little.

  1. fitted input called prediction [Appendix C.3 (Table 6) and Section 5.2.1 (Table 1)]
    "Default HPs are in bold. Since tCs < tCb, this setting totally generates 30 combinations of HPs. The result is in Table 6. ... VanillaAE GradAE (Ours) EntropyAE RandNet ROBOD AUC 0.758±0.004 0.775±0.003 0.769±0.005 0.736±0.00 0.744±0.00"

    The final reported GradAE AUC (0.775) is the control-variable performance of the default HP configuration chosen from a grid search conducted on the same 47 datasets that are then used to report the improvement over VanillaAE (0.758). In Table 6, the defaults (k=20, tD=1.57, [tCs,tCb]=[0.01,0.05]) are the best-performing settings, so the headline improvement is partly an artifact of selecting hyperparameters with access to the target AUC values. The HP-grid mean (0.765) still exceeds VanillaAE, so the result is not fully forced, but the published number is fitted to the evaluation benchmark rather than an out-of-sample prediction.

full rationale

The derivation chain in Section 4 is not definitionally circular: the cohesion metric C(G)=||sum g_i||/sum||g_i|| and the divergence metric D(G1,G2) are not defined in terms of AUC, and the stopping rule does not read labels. Theorem 4.1 gives a sufficient condition for inlier priority strengthening, and the connection to the algorithm is an admitted approximation, with the paper stating 'We left exploring more elaborate metric designs being closer to the theoretical conclusion as our future work'; this weakens the theory-to-algorithm bridge but does not make the theorem identical to the method. The main circularity is evaluational: the same 47 benchmark datasets are used both to select default hyperparameters and to report the final improvement, so the headline 0.758-to-0.775 gain is partially fitted to the test set. This is mitigated by the HP-grid mean also improving and by the method being label-free, but it prevents the reported numbers from being a clean out-of-sample prediction. The paper's own ablation (Table 6) shows tD=infty gives 0.760 versus 0.775 with tD=pi/2, so the advertised per-epoch cohesion mechanism is not the main driver of the reported gain; this is a correctness/evidence concern rather than a circular step. The only self-citation (EntropyStop's Rdown and patience mechanism) is not load-bearing because EntropyStop is an externally published method with its own benchmarks, and the novelty in GradStop is in the new metrics, not in the borrowed sliding-window early-stopping shell.

Assumptions & free parameters 5 free parameters · 5 assumptions · 0 invented entities

The central empirical result rests on a small set of algorithmic hyperparameters and on the heuristic that gradient norm ranking reflects inlier/outlier membership. No new physical entities are introduced.

free parameters (5)
  • k (GradSample sample size) = 20 for AE, DeepSVDD, RDP; 10 for VAE
    Number of gradient vectors selected as top-k and bottom-k; tuned via grid search in Appendix C.3.
  • tCs and tCb (cohesion thresholds) = [0.01, 0.05] for AE; [0.0, 0.1] for DeepSVDD; [0, 0.5] for RDP; [0.01, 0.5] for VAE
    Thresholds for |C_delta|; selected by grid search on the 47 datasets.
  • tD (divergence threshold) = 1.57 (pi/2) for AE and DeepSVDD; infinity for RDP and VAE
    Used to abort training if inlier and outlier gradient directions diverge too much; tuned in Table 6.
  • w (sliding window size) = 20 AE, 10 DeepSVDD, 50 RDP, 20 VAE
    Patience window; similar to EntropyStop's patience set to 20.
  • Rdown (downtrend threshold) = 0.001
    Borrowed from EntropyStop; fixed across models.
assumptions (5)
  • domain assumption Inlier priority holds: during training, average inlier losses decrease faster than average outlier losses.
    Stated in Section 1.2 and Appendix B; it is the core UOD assumption the method monitors.
  • domain assumption Gradient magnitudes of outliers are typically larger than those of inliers, and inlier gradients are more cohesive.
    Empirically observed in Section 4.1 and Figure 3; motivates GradSample.
  • standard math L-smoothness of per-class losses and a sufficiently small learning rate.
    Assumptions B.1 and B.2 in Appendix B for the theoretical bound.
  • ad hoc to paper The top-k and bottom-k gradient norm samples approximate the outlier and inlier sets at the collective level.
    Core assumption of GradSample (Section 4.2); the paper admits failure cases in Appendix C.5.
  • ad hoc to paper C(Glast)-C(Gtop) tracks log(r_t) from Theorem 4.1.
    Stated without proof at the end of Appendix B; bridges theory to the implemented metric.

how reviews work

0 comments
Cite this review

Pith. "Pith review of GradStop: Exploring Training Dynamics in Unsupervised Outlier Detection through Gradient." pith.science (2026). https://pith.science/paper/BGF4GJMB

@misc{pith2026241208501,
  author       = {Pith},
  title        = {Pith review of: GradStop: Exploring Training Dynamics in Unsupervised Outlier Detection through Gradient},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/BGF4GJMB}},
  note         = {Machine review of arXiv:2412.08501}
}
read the original abstract

Unsupervised Outlier Detection (UOD) is a critical task in data mining and machine learning, aiming to identify instances that significantly deviate from the majority. Without any label, deep UOD methods struggle with the misalignment between the model's direct optimization goal and the final performance goal of Outlier Detection (OD) task. Through the perspective of training dynamics, this paper proposes an early stopping algorithm to optimize the training of deep UOD models, ensuring they perform optimally in OD rather than overfitting the entire contaminated dataset. Inspired by UOD mechanism and inlier priority phenomenon, where intuitively models fit inliers more quickly than outliers, we propose GradStop, a sampling-based label-free algorithm to estimate model's real-time performance during training. First, a sampling method generates two sets: one likely containing more outliers and the other more inliers, then a metric based on gradient cohesion is applied to probe into current training dynamics, which reflects model's performance on OD task. Experimental results on 4 deep UOD algorithms and 47 real-world datasets and theoretical proofs demonstrate the effectiveness of our proposed early stopping algorithm in enhancing the performance of deep UOD models. Auto Encoder (AE) enhanced by GradStop achieves better performance than itself, other SOTA UOD methods, and even ensemble AEs. Our method provides a robust and effective solution to the problem of performance degradation during training, enabling deep UOD models to achieve better potential in anomaly detection tasks.

Figures

Figures reproduced from arXiv: 2412.08501 by the authors.

Figure 1
Figure 1. Performance degradation: UOD train￾ing process of AutoEncoder on clean dataset shuttle and original polluted shuttle. 0 20 40 60 80 100 0.7 0.8 0.9 AUC AE on vowels AUC Best AUC 0 20 40 60 80 100 Training iterations 5 10 Loss Train Loss Val Loss 0 20 40 60 80 100 0.55 0.60 0.65 AUC DeepSVDD on vowels AUC Best AUC 0 20 40 60 80 100 Training iterations 0.0 0.5 Loss Train Loss Val Loss [PITH_FULL_IMAGE:figures/full_fi… view at source ↗
Figure 3
Figure 3. Training dynamics of AE training on dataset [PITH_FULL_IMAGE:figures/full_fig_p005_3.png] view at source ↗
Figure 4
Figure 4. Case studies on datasets glass, wine, and optdigits, showing close correspondence between the C∆ and the variation trend of AUC. Top: AUC. Middle: C(Glast) and C(Gtop). Bottom: C ∆ = C(Glast) − C(Gtop). 22 [PITH_FULL_IMAGE:figures/full_fig_p022_4.png] view at source ↗
Figures from the paper (6 more)
Figure 5
Figure 5. Figure 5: AE: AUC curves vs. C∆ curves. Top: AUC. Middle: C(Glast) and C(Gtop). Bottom: C ∆ = C(Glast) − C(Gtop). 24 [PITH_FULL_IMAGE:figures/full_fig_p024_5.png]
Figure 6
Figure 6. Figure 6: AE: AUC curves vs. C∆ curves. Top: AUC. Middle: C(Glast) and C(Gtop). Bottom: C ∆ = C(Glast) − C(Gtop). 25 [PITH_FULL_IMAGE:figures/full_fig_p025_6.png]
Figure 7
Figure 7. Figure 7: AE: AUC curves vs. C∆ curves. Top: AUC. Middle: C(Glast) and C(Gtop). Bottom: C ∆ = C(Glast) − C(Gtop). 26 [PITH_FULL_IMAGE:figures/full_fig_p026_7.png]
Figure 8
Figure 8. Figure 8: AE: AUC curves vs. C∆ curves. Top: AUC. Middle: C(Glast) and C(Gtop). Bottom: C ∆ = C(Glast) − C(Gtop). 27 [PITH_FULL_IMAGE:figures/full_fig_p027_8.png]
Figure 9
Figure 9. Figure 9: AE: AUC curves vs. C∆ curves. Top: AUC. Middle: C(Glast) and C(Gtop). Bottom: C ∆ = C(Glast) − C(Gtop). 28 [PITH_FULL_IMAGE:figures/full_fig_p028_9.png]
Figure 10
Figure 10. Figure 10: AE: AUC curves vs. C∆ curves. Top: AUC. Middle: C(Glast) and C(Gtop). Bottom: C ∆ = C(Glast) − C(Gtop) [PITH_FULL_IMAGE:figures/full_fig_p029_10.png]

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

41 extracted references · 31 canonical work pages

  1. [1]

    S., Maharaj, T., Fischer, A., Courville, A., Bengio, Y., et al

    Arpit, D., Jastrz e bski, S., Ballas, N., Krueger, D., Bengio, E., Kanwal, M. S., Maharaj, T., Fischer, A., Courville, A., Bengio, Y., et al. A closer look at memorization in deep networks. In International conference on machine learning, pp.\ 233--242. PMLR, 2017

    2017

  2. [2]

    Understanding and improving early stopping for learning with noisy labels

    Bai, Y., Yang, E., Han, B., Yang, Y., Li, J., Mao, Y., Niu, G., and Liu, T. Understanding and improving early stopping for learning with noisy labels. In Advances in Neural Information Processing Systems, volume 34, pp.\ 24392--24403, 2021

    work page 2021

  3. [3]

    Bradley, A. P. The use of the area under the roc curve in the evaluation of machine learning algorithms. Pattern recognition, 30 0 (7): 0 1145--1159, 1997

    work page 1997

  4. [4]

    M., Kriegel, H.-P., Ng, R

    Breunig, M. M., Kriegel, H.-P., Ng, R. T., and Sander, J. Lof: identifying density-based local outliers. In Proceedings of the 2000 ACM SIGMOD international conference on Management of data, pp.\ 93--104, 2000

    work page 2000

  5. [5]

    and Chawla, S

    Chalapathy, R. and Chawla, S. Deep learning for anomaly detection: A survey. arXiv preprint arXiv:1901.03407, 2019

    arXiv 1901

  6. [6]

    Anomaly detection: A survey

    Chandola, V., Banerjee, A., and Kumar, V. Anomaly detection: A survey. ACM computing surveys (CSUR), 41 0 (3): 0 1--58, 2009

    work page 2009

  7. [7]

    Outlier detection with autoencoder ensembles

    Chen, J., Sathe, S., Aggarwal, C., and Turaga, D. Outlier detection with autoencoder ensembles. In Proceedings of the 2017 SIAM international conference on data mining, pp.\ 90--98. SIAM, 2017

    work page 2017

  8. [8]

    Hyperparameter sensitivity in deep outlier detection analysis and a scalable hyper-ensemble solution

    Ding, X., Zhao, L., and Akoglu, L. Hyperparameter sensitivity in deep outlier detection analysis and a scalable hyper-ensemble solution. In Proceedings of the 36th International Conference on Neural Information Processing Systems, NIPS '22, 2024

    work page 2024

Show all 41 references
  1. [9]

    Dou, Y., Liu, Z., Sun, L., Deng, Y., Peng, H., and Yu, P. S. Enhancing graph neural network-based fraud detectors against camouflaged fraudsters. In Proceedings of the 29th ACM International Conference on Information & Knowledge Management, pp.\ 315--324, 2020

    2020

  2. [10]

    A theoretical analysis of the learning dynamics under class imbalance

    Francazi, E., Baity-Jesi, M., and Lucchi, A. A theoretical analysis of the learning dynamics under class imbalance. In Krause, A., Brunskill, E., Cho, K., Engelhardt, B., Sabato, S., and Scarlett, J. (eds.), Proceedings of the 40th International Conference on Machine Learning,...

    2023

  3. [11]

    Adbench: Anomaly detection benchmark

    Han, S., Hu, X., Huang, H., Jiang, M., and Zhao, Y. Adbench: Anomaly detection benchmark. arXiv preprint arXiv:2206.09426, 2022

    2022 arXiv

  4. [12]

    Hawkins, D. M. Identification of outliers, volume 11. Springer, 1980

    1980

  5. [13]

    Discovering cluster-based local outliers

    He, Z., Xu, X., and Deng, S. Discovering cluster-based local outliers. Pattern Recogn. Lett., 24 0 (9–10): 0 1641–1650, 2003

    2003

  6. [14]

    Entropystop: Unsupervised deep outlier detection with loss entropy

    Huang, Y., Zhang, Y., Wang, L., Zhang, F., and Lin, X. Entropystop: Unsupervised deep outlier detection with loss entropy. In Proceedings of the 30th ACM SIGKDD Conference on Knowledge Discovery and Data Mining, 2024

    2024

  7. [15]

    Kiani, R., Jin, W., and Sheng, V. S. Survey on extreme learning machines for outlier detection. Mach. Learn., 2024

    2024

  8. [16]

    Odim: outlier detection via likelihood of under-fitted generative models

    Kim, D., Hwang, J., Lee, J., Kim, K., and Kim, Y. Odim: outlier detection via likelihood of under-fitted generative models. In Proceedings of the 41st International Conference on Machine Learning, ICML'24. JMLR.org, 2024

    2024

  9. [17]

    Kingma, D. P. and Welling, M. Auto-encoding variational bayes, 2022. URL https://arxiv.org/abs/1312.6114

    2022 arXiv

  10. [18]

    Revisiting time series outlier detection: Definitions and benchmarks

    Lai, K.-H., Zha, D., Xu, J., Zhao, Y., Wang, G., and Hu, X. Revisiting time series outlier detection: Definitions and benchmarks. In Thirty-fifth Conference on Neural Information Processing Systems Datasets and Benchmarks Track (Round 1), 2021

    2021

  11. [19]

    Gradient descent with early stopping is provably robust to label noise for overparameterized neural networks

    Li, M., Soltanolkotabi, M., and Oymak, S. Gradient descent with early stopping is provably robust to label noise for overparameterized neural networks. In International conference on artificial intelligence and statistics, pp.\ 4313--4324. PMLR, 2020

    2020

  12. [20]

    Ecod: Unsupervised outlier detection using empirical cumulative distribution functions

    Li, Z., Zhao, Y., Hu, X., Botta, N., Ionescu, C., and Chen, G. Ecod: Unsupervised outlier detection using empirical cumulative distribution functions. IEEE Transactions on Knowledge and Data Engineering, 2022

    2022

  13. [21]

    T., Ting, K

    Liu, F. T., Ting, K. M., and Zhou, Z.-H. Isolation forest. In 2008 eighth ieee international conference on data mining, pp.\ 413--422. IEEE, 2008

    2008

  14. [22]

    Bond: Benchmarking unsupervised outlier node detection on static attributed graphs

    Liu, K., Dou, Y., Zhao, Y., Ding, X., Hu, X., Zhang, R., Ding, K., Chen, C., Peng, H., Shu, K., et al. Bond: Benchmarking unsupervised outlier node detection on static attributed graphs. In Thirty-sixth Conference on Neural Information Processing Systems Datasets and Benchmark...

    2022

  15. [23]

    Generative adversarial active learning for unsupervised outlier detection

    Liu, Y., Li, Z., Zhou, C., Jiang, Y., Sun, J., Wang, M., and He, X. Generative adversarial active learning for unsupervised outlier detection. IEEE Transactions on Knowledge and Data Engineering, 32 0 (8): 0 1517--1528, 2019

    2019

  16. [24]

    and Bourlard, H

    Morgan, N. and Bourlard, H. Generalization and parameter estimation in feedforward nets: Some experiments. In Touretzky, D. (ed.), Advances in Neural Information Processing Systems, volume 2. Morgan-Kaufmann, 1989

    1989

  17. [25]

    Pang, G., Shen, C., Cao, L., and Hengel, A. V. D. Deep learning for anomaly detection: A review. ACM Computing Surveys (CSUR), 54 0 (2): 0 1--38, 2021

    2021

  18. [26]

    Neural transformation learning for deep anomaly detection beyond images

    Qiu, C., Pfrommer, T., Kloft, M., Mandt, S., and Rudolph, M. Neural transformation learning for deep anomaly detection beyond images. In International Conference on Machine Learning, pp.\ 8703--8714. PMLR, 2021

    2021

  19. [27]

    Latent outlier exposure for anomaly detection with contaminated data

    Qiu, C., Li, A., Kloft, M., Rudolph, M., and Mandt, S. Latent outlier exposure for anomaly detection with contaminated data. In International Conference on Machine Learning, pp.\ 18153--18167. PMLR, 2022

    2022

  20. [28]

    Efficient algorithms for mining outliers from large data sets

    Ramaswamy, S., Rastogi, R., and Shim, K. Efficient algorithms for mining outliers from large data sets. In Proceedings of the 2000 ACM SIGMOD international conference on Management of data, pp.\ 427--438, 2000

    2000

  21. [29]

    A., Binder, A., M \"u ller, E., and Kloft, M

    Ruff, L., Vandermeulen, R., Goernitz, N., Deecke, L., Siddiqui, S. A., Binder, A., M \"u ller, E., and Kloft, M. Deep one-class classification. In International conference on machine learning, pp.\ 4393--4402. PMLR, 2018

    2018

  22. [30]

    R., Vandermeulen, R

    Ruff, L., Kauffmann, J. R., Vandermeulen, R. A., Montavon, G., Samek, W., Kloft, M., Dietterich, T. G., and M \"u ller, K.-R. A unifying review of deep and shallow anomaly detection. Proceedings of the IEEE, 109 0 (5): 0 756--795, 2021

    2021

  23. [31]

    M., Schmidt-Erfurth, U., and Langs, G

    Schlegl, T., Seeb \"o ck, P., Waldstein, S. M., Schmidt-Erfurth, U., and Langs, G. Unsupervised anomaly detection with generative adversarial networks to guide marker discovery. In International conference on information processing in medical imaging, pp.\ 146--157. Springer, 2017

    2017

  24. [32]

    and Wolf, L

    Shenkar, T. and Wolf, L. Anomaly detection for tabular data with internal contrastive learning. In International Conference on Learning Representations, 2021

    2021

  25. [33]

    Unsupervised representation learning by predicting random distances

    Wang, H., Pang, G., Shen, C., and Ma, C. Unsupervised representation learning by predicting random distances. arXiv preprint arXiv:1912.12186, 2019 a

    1912 arXiv

  26. [34]

    Effective end-to-end unsupervised outlier detection via inlier priority of discriminative network

    Wang, S., Zeng, Y., Liu, X., Zhu, E., Yin, J., Xu, C., and Kloft, M. Effective end-to-end unsupervised outlier detection via inlier priority of discriminative network. Advances in neural information processing systems, 32, 2019 b

    2019

  27. [35]

    J., Borghetti, B

    Weller-Fahy, D. J., Borghetti, B. J., and Sodemann, A. A. A survey of distance and similarity measures used within network intrusion anomaly detection. IEEE Communications Surveys & Tutorials, 17 0 (1): 0 70--91, 2014

    2014

  28. [36]

    Robust early-learning: Hindering the memorization of noisy labels

    Xia, X., Liu, T., Han, B., Gong, C., Wang, N., Ge, Z., and Chang, Y. Robust early-learning: Hindering the memorization of noisy labels. In International conference on learning representations, 2020

    2020

  29. [37]

    Learning discriminative reconstructions for unsupervised outlier removal

    Xia, Y., Cao, X., Wen, F., Hua, G., and Sun, J. Learning discriminative reconstructions for unsupervised outlier removal. In Proceedings of the IEEE international conference on computer vision, pp.\ 1511--1519, 2015

    2015

  30. [38]

    O., Lee, C.-Y., and Pfister, T

    Yoon, J., Sohn, K., Li, C.-L., Arik, S. O., Lee, C.-Y., and Pfister, T. Self-trained one-class classification for unsupervised anomaly detection. arXiv e-prints, pp.\ arXiv--2106, 2021

    2021

  31. [39]

    and Paffenroth, R

    Zhou, C. and Paffenroth, R. C. Anomaly detection with robust deep autoencoders. In Proceedings of the 23rd ACM SIGKDD international conference on knowledge discovery and data mining, pp.\ 665--674, 2017

    2017

  32. [40]

    R., Cheng, W., Lumezanu, C., Cho, D., and Chen, H

    Zong, B., Song, Q., Min, M. R., Cheng, W., Lumezanu, C., Cho, D., and Chen, H. Deep autoencoding gaussian mixture model for unsupervised anomaly detection. In International conference on learning representations, 2018

    2018

  33. [41]

    write newline

    " write newline "" before.all 'output.state := FUNCTION n.dashify 't := "" t empty not t #1 #1 substring "-" = t #1 #2 substring "--" = not "--" * t #2 global.max substring 't := t #1 #1 substring "-" = "-" * t #2 global.max substring 't := while if t #1 #1 substring * t #2 gl...

Pith tools

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