pith. sign in

arxiv: 2603.23868 · v2 · submitted 2026-03-25 · 💻 cs.CV

MLE-UVAD: Minimal Latent Entropy Autoencoder for Fully Unsupervised Video Anomaly Detection

Yuang Geng , Junkai Zhou , Kang Yang , Pan He , Zhuoyang Zhou , Jose C. Principe , Joel Harley , Ivan Ruchkin This is my paper

Pith reviewed 2026-05-15 01:16 UTC · model grok-4.3

classification 💻 cs.CV
keywords unsupervised video anomaly detectionautoencoderlatent entropy minimizationreconstruction errorfully unsupervised
0
0 comments X

The pith

Adding minimal latent entropy loss to autoencoders creates a reconstruction gap for unsupervised video anomaly detection.

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

The paper introduces an autoencoder trained with both reconstruction loss and a minimal latent entropy loss on raw videos that include both normal and abnormal events. The entropy loss minimizes the uncertainty in the latent space, causing embeddings to cluster in high-density areas. Because normal frames are the majority in the video, anomalous embeddings are drawn into this normal cluster. Consequently, the decoder learns to reconstruct normal patterns accurately while struggling with anomalies, producing a detectable reconstruction error difference. This enables anomaly detection without any supervision or separate normal training data.

Core claim

The dual-loss design of reconstruction plus minimal latent entropy produces a clear reconstruction gap that enables effective anomaly detection in single-scene fully unsupervised video anomaly detection by pulling sparse anomalous latent embeddings into the dominant normal cluster.

What carries the argument

Minimal Latent Entropy (MLE) loss that minimizes the entropy of latent embeddings to encourage concentration around high-density regions.

If this is right

  • The approach outperforms baselines on two standard benchmarks and a self-collected driving dataset.
  • It works directly on raw videos without labels or normal-only data, avoiding issues with distribution shifts.
  • Anomaly detection is performed by thresholding the reconstruction error.
  • The method is robust in single-scene settings where anomalies are sparse.

Where Pith is reading between the lines

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

  • This technique might apply to other unsupervised outlier detection tasks in sequences where one class dominates.
  • If the proportion of anomalies increases, the clustering effect could weaken, requiring adjustments to the entropy term.
  • It implicitly performs a form of density estimation in latent space without explicit modeling.

Load-bearing premise

Normal frames dominate the raw video so that minimizing latent entropy pulls sparse anomalous embeddings into the normal cluster without any labels.

What would settle it

Evaluating the method on a dataset where anomalous frames constitute the majority would likely show degraded performance due to the loss of the dominance assumption.

Figures

Figures reproduced from arXiv: 2603.23868 by Ivan Ruchkin, Joel Harley, Jose C. Principe, Junkai Zhou, Kang Yang, Pan He, Yuang Geng, Zhuoyang Zhou.

Figure 1
Figure 1. Figure 1: Collapse of the anomalous latent distribution. Green: normal frames. Red: ab￾normal frames. Solid lines: latent distributions of the original inputs. Dashed lines: latent distributions of the reconstructions. The key idea is to pull sparse anomalous embeddings toward the dense normal embeddings, so normal frames remain well re￾constructed while anomalies reconstruct poorly. Our key novelty is that we colla… view at source ↗
Figure 2
Figure 2. Figure 2: MLE-UVAD pipeline: unsupervised training and detection through reconstruc￾tion gap. Unlabeled videos are directly put into an autoencoder trained with a dual loss: (1) The MSE loss reconstructs all frames accurately. (2) The MLE collapses sparse abnormal embeddings toward the dense normal embedding, making anomaly recon￾structions worse relative to normals. An anomaly is detected by setting a threshold on … view at source ↗
Figure 3
Figure 3. Figure 3: t-SNE visualization of latent embeddings across different methods (blue = normal, orange = abnormal). Baselines, only MSE and GCL [32], keep the normal and abnormal embeddings separate for better reconstruction. In contrast, our proposed MLE loss regularizes the latent distribution by collapsing the abnormal distribution into the dominant normal cluster. As shown in [PITH_FULL_IMAGE:figures/full_fig_p007_3.png] view at source ↗
Figure 4
Figure 4. Figure 4: Unsupervised anomaly detection performance across three benchmarks. Top row: baselines (TMAE, GCL, Vanilla CAE). Bottom row: our MLE-Guided CAE. Columns: DonkeyCar, UBnormal, Corridor. Our method shows a clear normal–anomaly separation of the PCC value. In contrast, the baselines fail to separate anomalies because they do not induce a reconstruction drop. Vanilla autoencoder, which minimizes global re￾cons… view at source ↗
Figure 5
Figure 5. Figure 5: Effect of kernel size σ in the MLE loss. Left: PCC trajectories for different σ values; shaded gray regions indicate ground-truth anomalies. Right: AUC heatmap across epochs. Mid-range σ (0.01–0.1) yields stable reconstructions and high AUC, while very small or large σ degrades detection. Gaussian affinities for embeddings, w_{ij} \;=\; \exp \!\Big (-\tfrac {(z_i - z_j)^2}{4\sigma ^2}\Big ), \label {eq:pai… view at source ↗
Figure 6
Figure 6. Figure 6: Effect of MLE loss weight λ on anomaly detection. Left: PCC trajectories for different λ values; shaded gray regions indicate ground-truth anomalies. Right: AUC heatmap across epochs. Moderate λ (0.1–1.0) achieves the best trade-off between stable reconstruction and clear anomaly sensitivity. high AUC across epochs. In short, λ values from 0.001 to 1 and σ values from 0.01 to 0.1 consistently achieve near-… view at source ↗
Figure 7
Figure 7. Figure 7: Impact of anomaly ratios on detection performance for the Corridor dataset. The subplots show the PCC trajectories at different anomaly ratios, with the gray regions indicating anomalies. Effect of the Anomaly Ratio on Detection Performance. Furthermore, we evaluated our model’s performance across anomaly ratios ranging from 10% to 60% on all three datasets. When the anomaly ratio is below 40%, the model m… view at source ↗
Figure 8
Figure 8. Figure 8: Test of generalization ability across 10 different anomalies on the Corridor dataset. The model is trained on normal data with only protest abnormal data (gray area: protest) and is tested on the other 10 anomalies. Generalization Capability Across Different Types of Anomalies. We evaluate the generalization ability by training our model on an unlabeled video containing normal scenes and a single anomaly t… view at source ↗
read the original abstract

In this paper, we address the challenging problem of single-scene, fully unsupervised video anomaly detection (VAD), where raw videos containing both normal and abnormal events are used directly for training and testing without any labels. This differs sharply from prior work that either requires extensive labeling (fully or weakly supervised) or depends on normal-only videos (one-class classification), which are vulnerable to distribution shifts and contamination. We propose an entropy-guided autoencoder that detects anomalies through reconstruction error by reconstructing normal frames well while making anomalies reconstruct poorly. The key idea is to combine the standard reconstruction loss with a novel Minimal Latent Entropy (MLE) loss in the autoencoder. Reconstruction loss alone maps normal and abnormal inputs to distinct latent clusters due to their inherent differences, but also risks reconstructing anomalies too well to detect. Therefore, MLE loss addresses this by minimizing the entropy of latent embeddings, encouraging them to concentrate around high-density regions. Since normal frames dominate the raw video, sparse anomalous embeddings are pulled into the normal cluster, so the decoder emphasizes normal patterns and produces poor reconstructions for anomalies. This dual-loss design produces a clear reconstruction gap that enables effective anomaly detection. Extensive experiments on two widely used benchmarks and a challenging self-collected driving dataset demonstrate that our method achieves robust and superior performance over baselines.

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 MLE-UVAD, an autoencoder for fully unsupervised single-scene video anomaly detection that trains directly on raw mixed videos. It augments standard reconstruction loss with a Minimal Latent Entropy (MLE) loss that minimizes the entropy of latent embeddings. The central claim is that reconstruction loss alone separates normal and anomalous inputs into distinct clusters, while the MLE term concentrates embeddings around high-density regions; because normal frames dominate, sparse anomalous latents are pulled into the normal mode, causing the decoder to reconstruct normals well and anomalies poorly and thereby producing a usable reconstruction-error gap for detection. Experiments on two public benchmarks plus a self-collected driving dataset are reported to show superior performance over baselines.

Significance. If the claimed clustering mechanism is shown to hold, the method would constitute a meaningful advance for unsupervised VAD by removing dependence on labels or clean normal-only training sets and by offering robustness to distribution shift. The dual-loss formulation is conceptually simple and could be portable to other reconstruction-based anomaly tasks, provided the entropy term reliably produces the asserted normal-mode attractor rather than collapse or multi-modal equilibria.

major comments (3)
  1. [Abstract and §3] Abstract and §3: the claim that MLE loss 'pulls sparse anomalous embeddings into the normal cluster' because normals dominate is stated without derivation, gradient-flow analysis, or proof that the joint loss has a unique high-density normal attractor. No equation or lemma shows why the entropy term does not induce uniform collapse or preserve separate modes of comparable density, which is load-bearing for the reconstruction-gap argument.
  2. [Experiments] Experiments section: superior benchmark numbers are presented, yet no ablation isolates the MLE term from generic regularization, no latent-space diagnostics (t-SNE, entropy histograms, mode-separation metrics) are supplied, and no controlled variation of anomaly ratio is reported to test the dominance assumption. These omissions leave the central mechanism unverified.
  3. [§3] §3: the balancing weight between reconstruction and MLE losses is treated as a free hyper-parameter; its sensitivity and the range over which the claimed clustering occurs should be quantified, as the method's robustness claim rests on this choice.
minor comments (2)
  1. [Abstract] The abstract would be clearer if it briefly stated the mathematical form of the MLE loss (e.g., entropy estimator used) rather than describing it only at the level of 'minimizing entropy of latent embeddings.'
  2. [§3] Notation for the latent distribution and entropy estimator should be introduced once in §3 and used consistently thereafter to avoid ambiguity.

Simulated Author's Rebuttal

3 responses · 0 unresolved

We thank the referee for the constructive comments. We address each major point below and have revised the manuscript to incorporate additional analysis, ablations, and sensitivity studies to better substantiate our claims.

read point-by-point responses

  1. Referee: Abstract and §3: the claim that MLE loss 'pulls sparse anomalous embeddings into the normal cluster' because normals dominate is stated without derivation, gradient-flow analysis, or proof that the joint loss has a unique high-density normal attractor. No equation or lemma shows why the entropy term does not induce uniform collapse or preserve separate modes of comparable density, which is load-bearing for the reconstruction-gap argument.

    Authors: We agree that a more rigorous justification is needed. In the revised manuscript, we have added a gradient analysis in Section 3.2 showing that the MLE loss term creates an attractive force towards high-density regions in latent space. Combined with the reconstruction loss, which preserves separation between normal and anomalous modes, the joint optimization leads to anomalous points being pulled into the dominant normal mode. We include a simple derivation demonstrating that uniform collapse is prevented by the reconstruction term maintaining distinct reconstruction errors. Empirical support is provided via new latent visualizations. revision: yes

  2. Referee: Experiments section: superior benchmark numbers are presented, yet no ablation isolates the MLE term from generic regularization, no latent-space diagnostics (t-SNE, entropy histograms, mode-separation metrics) are supplied, and no controlled variation of anomaly ratio is reported to test the dominance assumption. These omissions leave the central mechanism unverified.

    Authors: We have revised the Experiments section to include comprehensive ablations. Specifically, we compare against variants using L2 regularization or dropout instead of MLE to isolate its effect. We provide t-SNE plots of latent embeddings, histograms of latent entropy, and quantitative mode separation metrics (e.g., silhouette score). Additionally, we report results with anomaly ratios varied from 5% to 30% in synthetic mixtures, confirming the dominance assumption holds and performance degrades gracefully as anomaly ratio increases. revision: yes

  3. Referee: §3: the balancing weight between reconstruction and MLE losses is treated as a free hyper-parameter; its sensitivity and the range over which the claimed clustering occurs should be quantified, as the method's robustness claim rests on this choice.

    Authors: We have added a sensitivity analysis for the balancing hyper-parameter λ in the revised Section 3 and Experiments. We evaluate performance for λ ranging from 0.01 to 5.0 on the benchmarks, showing stable superior results for λ ∈ [0.1, 1.0]. Outside this range, either the reconstruction gap diminishes (low λ) or training becomes unstable (high λ). We recommend λ=0.5 as default and discuss how to tune it based on dataset characteristics. revision: yes

Circularity Check

0 steps flagged

No circularity in the proposed dual-loss autoencoder design

full rationale

The paper proposes a new Minimal Latent Entropy (MLE) loss term added to standard reconstruction loss. The central mechanism—that entropy minimization concentrates embeddings and pulls sparse anomalies into the dominant normal cluster—is presented as an intuitive consequence of data dominance and the joint loss, without any equation that reduces the claimed reconstruction gap to a fitted parameter or self-referential definition. No self-citations, uniqueness theorems, or ansatzes from prior author work are invoked as load-bearing steps. The derivation chain is self-contained as a design choice whose validity rests on empirical benchmark results rather than circular reduction.

Axiom & Free-Parameter Ledger

1 free parameters · 1 axioms · 0 invented entities

The central claim rests on the domain assumption that normal frames dominate the data distribution and that entropy minimization will therefore cluster anomalies with normals; no free parameters or invented entities are explicitly introduced in the abstract.

free parameters (1)
  • balancing weight between reconstruction and MLE losses
    A scalar hyperparameter must be chosen to trade off the two terms; its value is not stated in the abstract.
axioms (1)
  • domain assumption Normal frames dominate the raw video
    Explicitly invoked to justify why anomalous embeddings are pulled into the normal cluster.

pith-pipeline@v0.9.0 · 5552 in / 1209 out tokens · 48897 ms · 2026-05-15T01:16:26.668915+00:00 · methodology

discussion (0)

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

Forward citations

Cited by 1 Pith paper

Reviewed papers in the Pith corpus that reference this work. Sorted by Pith novelty score.

  1. COPRA: Conditional Parameter Adaptation with Reinforcement Learning for Video Anomaly Detection

    cs.CV 2026-05 unverdicted novelty 5.0

    COPRA introduces conditional parameter adaptation via RL to dynamically tune frozen VLMs for video anomaly detection, outperforming static methods in in-domain and cross-domain settings while generalizing to other vid...

Reference graph

Works this paper leans on

38 extracted references · 38 canonical work pages · cited by 1 Pith paper

  1. [1]

    In: Proceedings of the IEEE/CVF conference on computer vision and pattern recognition

    Acsintoae,A.,Florescu,A.,Georgescu,M.I.,Mare,T.,Sumedrea,P.,Ionescu,R.T., Khan, F.S., Shah, M.: Ubnormal: New benchmark for supervised open-set video anomaly detection. In: Proceedings of the IEEE/CVF conference on computer vision and pattern recognition. pp. 20143–20153 (2022)

    work page 2022

  2. [2]

    In: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR)

    Al-lahham, A., Zaheer, M.Z., Tastan, N., Nandakumar, K.: Collaborative learning of anomalies with privacy (clap) for unsupervised video anomaly detection: A new baseline. In: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR). pp. 12416–12425 (June 2024)

    work page 2024

  3. [3]

    IEEE Transactions on Systems, Man, and Cybernetics: Systems 51(9), 5819–5829 (2019)

    Chen, B., Dang, L., Gu, Y., Zheng, N., Príncipe, J.C.: Minimum error entropy kalman filter. IEEE Transactions on Systems, Man, and Cybernetics: Systems 51(9), 5819–5829 (2019)

    work page 2019

  4. [4]

    In: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) Workshops

    Dawoud, K., Zaheer, Z., Khan, M., Nandakumar, K., Elsaddik, A., Khan, M.H.: Fusedvision: A knowledge-infusing approach for practical anomaly detection in real-world surveillance videos. In: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) Workshops. pp. 4045–4055 (June 2025)

    work page 2025

  5. [5]

    In: The Twelfth International Conference on Learning Representations (2024)

    Dong, Y., Gong, T., Chen, H., Yu, S., Li, C.: Rethinking information-theoretic generalization: Loss entropy induced pac bounds. In: The Twelfth International Conference on Learning Representations (2024)

    work page 2024

  6. [6]

    In: 2025 IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS)

    Geng, Y., Zhou, Y., Zhang, Y., Zhang, Z.R., Yang, K., Ruble, T., Vidal, G., Ruchkin, I.: Unsupervised anomaly detection improves imitation learning for autonomous racing. In: 2025 IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS). pp. 11654–11660. IEEE (2025)

    work page 2025

  7. [7]

    In: 2022 IEEE International Conference on Multimedia and Expo (ICME)

    Hu,J.,Yu,G.,Wang,S.,Zhu,E.,Cai,Z.,Zhu,X.:Detectinganomalouseventsfrom unlabeled videos via temporal masked auto-encoding. In: 2022 IEEE International Conference on Multimedia and Expo (ICME). pp. 1–6. IEEE (2022)

    work page 2022

  8. [8]

    In: 2025 IEEE/CVF Winter Conference on Ap- plications of Computer Vision (WACV)

    Im, J., Son, Y., Hong, J.H.: Fun-ad: Fully unsupervised learning for anomaly de- tection with noisy training data. In: 2025 IEEE/CVF Winter Conference on Ap- plications of Computer Vision (WACV). pp. 9447–9456. IEEE (2025)

    work page 2025

  9. [9]

    In: Proceedings of the IEEE/CVF winter conference on applications of computer vision

    Karim, H., Doshi, K., Yilmaz, Y.: Real-time weakly supervised video anomaly detection. In: Proceedings of the IEEE/CVF winter conference on applications of computer vision. pp. 6848–6856 (2024)

    work page 2024

  10. [10]

    Geng et al

    Kommanduri, R., Ghorai, M.: Dast-net: Dense visual attention augmented spatio- temporalnetworkforunsupervisedvideoanomalydetection.Neurocomputing579, 127444 (2024) 16 Y. Geng et al

    work page 2024

  11. [11]

    In: Proceedings of the AAAI conference on artificial intelligence

    Li, H., Yu, S., Principe, J.: Causal recurrent variational autoencoder for medi- cal time series generation. In: Proceedings of the AAAI conference on artificial intelligence. vol. 37, pp. 8562–8570 (2023)

    work page 2023

  12. [12]

    Liu, F., Huang, X., Chen, Y., Suykens, J.A.: Random features for kernel approxi- mation:Asurveyonalgorithms,theory,andbeyond.IEEETransactionsonPattern Analysis and Machine Intelligence44(10), 7128–7148 (2021)

    work page 2021

  13. [13]

    ACM Comput

    Liu, J., Liu, Y., Lin, J., Li, J., Cao, L., Sun, P., Hu, B., Song, L., Boukerche, A., Leung, V.C.: Networking systems for video anomaly detection: A tutorial and survey. ACM Comput. Surv.57(10) (May 2025).https://doi.org/10.1145/ 3729222,https://doi.org/10.1145/3729222

    work page doi:10.1145/3729222 2025

  14. [14]

    IEEE Transactions on Neural Networks and Learning Systems pp

    Liu, Y., Liu, S., Zhu, X., Yang, H., Li, J., Guo, J., Teng, L., Yang, D., Wang, Y., Liu, J.: Privacy-preserving video anomaly detection: A survey. IEEE Transactions on Neural Networks and Learning Systems pp. 1–22 (2025).https://doi.org/10. 1109/TNNLS.2025.3600252

    work page arXiv 2025

  15. [15]

    In: 2025 IEEE 101st Vehicular Technology Conference (VTC2025-Spring)

    Pallewela, R., Eldeeb, E., Alves, H.: An analysis of minimum error entropy loss functions in wireless communications. In: 2025 IEEE 101st Vehicular Technology Conference (VTC2025-Spring). pp. 1–6. IEEE (2025)

    work page 2025

  16. [16]

    In: Proceedings of the IEEE/CVF conference on computer vision and pattern recognition

    Pang, G., Yan, C., Shen, C., Hengel, A.v.d., Bai, X.: Self-trained deep ordinal re- gression for end-to-end video anomaly detection. In: Proceedings of the IEEE/CVF conference on computer vision and pattern recognition. pp. 12173–12182 (2020)

    work page 2020

  17. [17]

    Pattern Recognition153, 110550 (2024)

    Qiu, S., Ye, J., Zhao, J., He, L., Liu, L., Huang, X., et al.: Video anomaly detection guided by clustering learning. Pattern Recognition153, 110550 (2024)

    work page 2024

  18. [18]

    IEEE Transactions on Pattern Analysis and Machine Intelli- gence44(5), 2293–2312 (2022).https://doi.org/10.1109/TPAMI.2020.3040591

    Ramachandra, B., Jones, M.J., Vatsavai, R.R.: A survey of single-scene video anomaly detection. IEEE Transactions on Pattern Analysis and Machine Intelli- gence44(5), 2293–2312 (2022).https://doi.org/10.1109/TPAMI.2020.3040591

    work page doi:10.1109/tpami.2020.3040591 2022

  19. [19]

    In: 2017 IEEE international conference on image processing (ICIP)

    Ravanbakhsh, M., Nabi, M., Sangineto, E., Marcenaro, L., Regazzoni, C., Sebe, N.: Abnormal event detection in videos using generative adversarial nets. In: 2017 IEEE international conference on image processing (ICIP). pp. 1577–1581. IEEE (2017)

    work page 2017

  20. [20]

    In: Proceedings of the fourth Berkeley symposium on mathematical statistics and probability, volume 1: con- tributions to the theory of statistics

    Rényi, A.: On measures of entropy and information. In: Proceedings of the fourth Berkeley symposium on mathematical statistics and probability, volume 1: con- tributions to the theory of statistics. vol. 4, pp. 547–562. University of California Press (1961)

    work page 1961

  21. [21]

    In: Proceedings of the IEEE/CVF winter conference on applications of computer vision

    Rodrigues, R., Bhargava, N., Velmurugan, R., Chaudhuri, S.: Multi-timescale tra- jectory prediction for abnormal human activity detection. In: Proceedings of the IEEE/CVF winter conference on applications of computer vision. pp. 2626–2634 (2020)

    work page 2020

  22. [22]

    In: Proceedings of the IEEE conference on computer vision and pattern recognition

    Sultani, W., Chen, C., Shah, M.: Real-world anomaly detection in surveillance videos. In: Proceedings of the IEEE conference on computer vision and pattern recognition. pp. 6479–6488 (2018)

    work page 2018

  23. [23]

    and WOLF, M

    Sun, Z., Wang, P., Zheng, W., Zhang, M.: Dual groupgan: An unsupervised four-competitor (2v2) approach for video anomaly detection. Pattern Recog- nition153, 110500 (2024).https://doi.org/https://doi.org/10.1016/j. patcog.2024.110500,https://www.sciencedirect.com/science/article/pii/ S0031320324002516

    work page doi:10.1016/j 2024

  24. [24]

    IEEE Transactions on Mul- timedia26, 10160–10173 (2024)

    Tao, C., Wang, C., Lin, S., Cai, S., Li, D., Qian, J.: Feature reconstruction with disruption for unsupervised video anomaly detection. IEEE Transactions on Mul- timedia26, 10160–10173 (2024)

    work page 2024

  25. [25]

    Tian, Y., Pang, G., Chen, Y., Singh, R., Verjans, J.W., Carneiro, G.: Weakly- supervised video anomaly detection with robust temporal feature magnitude learn- Title Suppressed Due to Excessive Length 17 ing.In:ProceedingsoftheIEEE/CVFinternationalconferenceoncomputervision. pp. 4975–4986 (2021)

    work page 2021

  26. [26]

    Scientific Reports15(1), 1–18 (2025)

    Verma, U., Pai, M.M.M., Pai, R.M., et al.: Contextual information based anomaly detection for multi-scene aerial videos. Scientific Reports15(1), 1–18 (2025)

    work page 2025

  27. [27]

    Mixing neural networks and the

    Viitala, A., Boney, R., Zhao, Y., Ilin, A., Kannala, J.: Learning to drive (l2d) as a low-cost benchmark for real-world reinforcement learning. In: 2021 20th International Conference on Advanced Robotics (ICAR). pp. 275–281 (2021). https://doi.org/10.1109/ICAR53236.2021.9659342

    work page doi:10.1109/icar53236.2021.9659342 2021

  28. [28]

    arXiv preprint arXiv:2104.07268 (2021)

    Wan, B., Fang, Y., Xia, X., Mei, J.: Weakly supervised video anomaly detection via center-guided discriminative learning. arXiv preprint arXiv:2104.07268 (2021)

    work page arXiv 2021

  29. [29]

    IEEE transactions on neural networks and learning systems33(6), 2301–2312 (2021)

    Wang, X., Che, Z., Jiang, B., Xiao, N., Yang, K., Tang, J., Ye, J., Wang, J., Qi, Q.: Robust unsupervised video anomaly detection by multipath frame predic- tion. IEEE transactions on neural networks and learning systems33(6), 2301–2312 (2021)

    work page 2021

  30. [30]

    IEEE Transactions on Industrial Informatics (2025)

    Yang, K., Lin, Z., Yang, Z., Tian, Z., Ma, J., Príncipe, J.C., Harley, J.B.: Im- proved pca reconstruction-based unsupervised anomaly detection in uncontrolled structural health monitoring with correntropy. IEEE Transactions on Industrial Informatics (2025)

    work page 2025

  31. [31]

    Journal of the Franklin Institute356(5), 3187–3215 (2019)

    Yu, S., Abraham, Z., Wang, H., Shah, M., Wei, Y., Príncipe, J.C.: Concept drift de- tection and adaptation with hierarchical hypothesis testing. Journal of the Franklin Institute356(5), 3187–3215 (2019)

    work page 2019

  32. [32]

    In: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR)

    Zaheer, M.Z., Mahmood, A., Khan, M.H., Segu, M., Yu, F., Lee, S.I.: Generative cooperative learning for unsupervised video anomaly detection. In: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR). pp. 14744–14754 (June 2022)

    work page 2022

  33. [33]

    In: European Conference on Computer Vision

    Zhang, G., Liu, Y., Yang, X., Huang, H., Huang, C.: Trafficnight: An aerial multi- modal benchmark for nighttime vehicle surveillance. In: European Conference on Computer Vision. pp. 36–48. Springer (2024)

    work page 2024

  34. [34]

    URL https://ieeexplore.ieee.org/document/ 10658325/

    Zhang, M., Wang, J., Qi, Q., Sun, H., Zhuang, Z., Ren, P., Ma, R., Liao, J.: Multi-scale video anomaly detection by multi-grained spatio-temporal represen- tation learning. In: 2024 IEEE/CVF Conference on Computer Vision and Pat- tern Recognition (CVPR). pp. 17385–17394 (2024).https://doi.org/10.1109/ CVPR52733.2024.01646

    work page arXiv 2024

  35. [35]

    In: Pro- ceedings of the IEEE/CVF conference on computer vision and pattern recognition

    Zhong, J.X., Li, N., Kong, W., Liu, S., Li, T.H., Li, G.: Graph convolutional label noise cleaner: Train a plug-and-play action classifier for anomaly detection. In: Pro- ceedings of the IEEE/CVF conference on computer vision and pattern recognition. pp. 1237–1246 (2019)

    work page 2019

  36. [36]

    In: Proceedings of the 23rd ACM SIGKDD International Conference on Knowledge Discovery and Data Mining

    Zhou, C., Paffenroth, R.C.: Anomaly detection with robust deep autoencoders. In: Proceedings of the 23rd ACM SIGKDD International Conference on Knowledge Discovery and Data Mining. p. 665–674. KDD ’17, Association for Computing Machinery, New York, NY, USA (2017).https://doi.org/10.1145/3097983. 3098052,https://doi.org/10.1145/3097983.3098052

    work page doi:10.1145/3097983 2017

  37. [37]

    In: Globerson, A., Mackey, L., Bel- grave, D., Fan, A., Paquet, U., Tomczak, J., Zhang, C

    Zhu, L., Wang, L., Raj, A., Gedeon, T., Chen, C.: Advancing video anomaly de- tection: A concise review and a new dataset. In: Globerson, A., Mackey, L., Bel- grave, D., Fan, A., Paquet, U., Tomczak, J., Zhang, C. (eds.) Advances in Neu- ral Information Processing Systems. vol. 37, pp. 89943–89977. Curran Associates, Inc. (2024),https://proceedings.neurip...

    work page 2024

  38. [38]

    In: International conference on learning representations (2018)

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

    work page 2018