pith. sign in

arxiv: 2606.29721 · v1 · pith:VE6KB5JQnew · submitted 2026-06-29 · 💻 cs.LG · cs.AI

Redefining Maritime Anomaly Detection via Equation-Grounded Synthetic Anomalies

Youngseok Hwang , Sungho Bae , Dohun Lee , Jaeeun Seo , Jeehong Kim , Wonhee Lee , Hyunwoo Park This is my paper

Pith reviewed 2026-06-30 07:48 UTC · model grok-4.3

classification 💻 cs.LG cs.AI
keywords maritime anomaly detectionAIS datasynthetic anomaliesanomaly taxonomyLLM-guided synthesistime-series modelsbenchmark evaluation
0
0 comments X

The pith

Equations define three maritime anomaly types to enable scalable labeled dataset creation from AIS data.

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

The paper proposes an equation-grounded anomaly taxonomy consisting of unexpected AIS activity, route deviation, and close approach to address limitations in existing maritime anomaly detection methods that rely on rarity or expert labeling. This taxonomy is designed to be implementable with limited AIS data and to capture interaction-driven hazards like near-misses. Building on it, the authors introduce a pipeline that uses LLM-guided plausibility scores to synthesize anomalies and generate timestamp-level labels. They also create benchmark settings to evaluate various models under different conditions. A sympathetic reader would care because this offers a systematic, scalable alternative for improving safety and traffic management at sea.

Core claim

The paper establishes an equation-grounded anomaly taxonomy with three types—unexpected AIS activity (A1), route deviation (A2), and close approach (A3)—that covers single-vessel and inter-vessel anomalies, and a unified score-synthesize-label pipeline that produces LLM-guided plausibility scores to synthesize anomalies and assign timestamp-level labels, providing a basis for evaluating detection methods across anomaly types and temporal windows.

What carries the argument

The equation-grounded anomaly taxonomy of types A1, A2, and A3, which provides implementable mathematical definitions for anomalies under limited AIS observation schema and supports the synthesis pipeline.

Load-bearing premise

The equations defining the three anomaly types accurately identify practically critical and interaction-driven hazards that prior methods miss, and the LLM-guided synthesis produces anomalies whose distribution supports valid model evaluation.

What would settle it

A direct comparison where models trained on the synthesized labels show no improvement or fail to detect known real-world near-miss incidents when tested against actual reported maritime events.

Figures

Figures reproduced from arXiv: 2606.29721 by Dohun Lee, Hyunwoo Park, Jaeeun Seo, Jeehong Kim, Sungho Bae, Wonhee Lee, Youngseok Hwang.

Figure 1
Figure 1. Figure 1: Motivation and Overview. Comparison between pre [PITH_FULL_IMAGE:figures/full_fig_p002_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: Overview of the framework. We propose a pipeline for synthesizing and labeling maritime anomalies in AIS data. An [PITH_FULL_IMAGE:figures/full_fig_p004_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: Equation-grounded anomaly examples for three [PITH_FULL_IMAGE:figures/full_fig_p005_3.png] view at source ↗
Figure 4
Figure 4. Figure 4: Distributional analysis of LLM scores across anom [PITH_FULL_IMAGE:figures/full_fig_p006_4.png] view at source ↗
Figure 5
Figure 5. Figure 5: Decision-level repeatability of the LLM scorer. Each [PITH_FULL_IMAGE:figures/full_fig_p006_5.png] view at source ↗
Figure 6
Figure 6. Figure 6: Comparison of sparsity-based and equation [PITH_FULL_IMAGE:figures/full_fig_p007_6.png] view at source ↗
read the original abstract

Maritime anomaly detection is essential for ensuring maritime safety, security, and efficient traffic management at sea, with Automatic Identification System (AIS) data serving as a primary data source. Despite its importance, most publicly available AIS datasets lack predefined anomaly labels, forcing prior studies to rely on either distribution-based rarity or domain rule/expert-assisted labeling. These approaches, however, face fundamental limitations: statistical rarity often fails to reflect practically critical events, while expert-based labeling is costly, subjective, and difficult to scale. Moreover, both paradigms tend to overlook interaction-driven hazards such as near-miss approaches between vessels. To address these challenges, we propose an equation-grounded anomaly taxonomy that is implementable under a limited AIS observation schema and extensible to other AIS datasets. Specifically, the taxonomy defines three anomaly types: unexpected AIS activity (A1), route deviation (A2), and close approach (A3), covering both single-vessel and inter-vessel anomalies. Building on this taxonomy, we introduce a unified score-synthesize-label pipeline that produces LLM-guided plausibility scores, uses them to synthesize anomalies, and assigns timestamp-level labels. To rigorously assess detection performance, we further design benchmark evaluation settings that account for variations in temporal-window length and anomaly-type composition, and evaluate a broad range of time-series models and anomaly detection models. Together, these contributions provide a systematic basis for evaluating maritime anomaly detection methods across different anomaly types. Our code is available at https://github.com/snudial/open-maritime-anomaly-detection.

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 manuscript proposes an equation-grounded anomaly taxonomy for maritime anomaly detection from limited AIS data, defining three types (A1: unexpected AIS activity, A2: route deviation, A3: close approach) that cover single- and inter-vessel cases. It introduces a unified score-synthesize-label pipeline that generates LLM-guided plausibility scores, synthesizes anomalies, and produces timestamp-level labels, together with benchmark settings that vary temporal windows and anomaly-type composition for evaluating time-series and anomaly detection models.

Significance. If the taxonomy equations and LLM synthesis are shown to align with real hazards, the work would supply a scalable, less subjective benchmark for maritime anomaly detection that targets interaction-driven events missed by rarity- or expert-based methods. The public code release is a concrete strength that enables direct reproducibility and extension.

major comments (3)
  1. [Anomaly Taxonomy Definitions] Anomaly taxonomy (A1–A3 definitions): the equations are simple threshold rules on distance/speed/route fields, yet no external validation against documented near-misses, COLREG violations, or expert labels is supplied. This directly undermines the central claim that the taxonomy identifies practically critical hazards missed by prior methods.
  2. [Score-Synthesize-Label Pipeline] Score-synthesize-label pipeline: because the LLM plausibility scores are derived from the same unvalidated equations, the synthesized anomalies inherit the same mapping problem; no analysis demonstrates that their distribution matches real hazard statistics or supports valid downstream evaluation.
  3. [Benchmark Evaluation] Benchmark evaluation settings: although the abstract states that models are evaluated across temporal-window lengths and anomaly-type compositions, the manuscript supplies no quantitative results, performance tables, or ablation on whether the synthetic labels produce meaningful detection rankings.
minor comments (2)
  1. Explicitly list the numerical thresholds and any free parameters used in the A1–A3 equations so readers can assess sensitivity.
  2. Clarify the precise AIS fields required by the limited observation schema and how the taxonomy remains extensible when additional fields become available.

Simulated Author's Rebuttal

3 responses · 0 unresolved

We thank the referee for the constructive and detailed comments. We address each major point below, clarifying the intended scope of the taxonomy and pipeline as a reproducible synthetic benchmark rather than a validated real-world hazard detector. Where appropriate, we indicate revisions to strengthen the manuscript.

read point-by-point responses

  1. Referee: [Anomaly Taxonomy Definitions] Anomaly taxonomy (A1–A3 definitions): the equations are simple threshold rules on distance/speed/route fields, yet no external validation against documented near-misses, COLREG violations, or expert labels is supplied. This directly undermines the central claim that the taxonomy identifies practically critical hazards missed by prior methods.

    Authors: The taxonomy is explicitly presented as an equation-grounded, implementable definition under limited AIS schemas to enable synthetic label generation where none exist, not as a claim of direct equivalence to real documented hazards. The abstract and introduction emphasize providing "a systematic basis for evaluating maritime anomaly detection methods across different anomaly types" rather than asserting that the thresholds match all COLREG violations or near-misses. We agree that stronger positioning is needed and will revise the introduction and a new limitations subsection to explicitly state that the equations are heuristic proxies derived from AIS fields and domain literature, without external validation, and to discuss the distinction between synthetic benchmarking and real-hazard alignment. revision: partial

  2. Referee: [Score-Synthesize-Label Pipeline] Score-synthesize-label pipeline: because the LLM plausibility scores are derived from the same unvalidated equations, the synthesized anomalies inherit the same mapping problem; no analysis demonstrates that their distribution matches real hazard statistics or supports valid downstream evaluation.

    Authors: The pipeline is designed to produce anomalies that are internally consistent with the defined taxonomy equations and to assign timestamp-level labels for controlled evaluation; it does not claim to reproduce the statistical distribution of real hazards, which would require unavailable ground-truth labels. We will add a new subsection in the experiments that reports basic statistics of the generated anomalies (e.g., frequency per type, temporal characteristics) and an explicit statement that downstream model rankings are meaningful only within the synthetic benchmark, not as proxies for real-world performance. revision: yes

  3. Referee: [Benchmark Evaluation] Benchmark evaluation settings: although the abstract states that models are evaluated across temporal-window lengths and anomaly-type compositions, the manuscript supplies no quantitative results, performance tables, or ablation on whether the synthetic labels produce meaningful detection rankings.

    Authors: The current manuscript describes the benchmark settings (temporal windows and anomaly-type compositions) and the models considered, but does not yet include the actual numerical results or tables. We will add a dedicated experimental section with performance tables, rankings across settings, and ablations on anomaly-type composition to demonstrate that the synthetic labels yield differentiated and interpretable detection outcomes. revision: yes

Circularity Check

0 steps flagged

No circularity detected in derivation chain

full rationale

The paper defines a new anomaly taxonomy (A1 unexpected AIS activity, A2 route deviation, A3 close approach) via implementable equations on limited AIS fields, then builds a score-synthesize-label pipeline around those definitions. No quoted step shows a result reducing by construction to its own inputs, fitted parameters renamed as predictions, or load-bearing self-citation chains. The central claims rest on the novelty of the taxonomy and pipeline rather than any self-referential equivalence.

Axiom & Free-Parameter Ledger

1 free parameters · 1 axioms · 1 invented entities

The approach rests on domain assumptions about AIS data fields being sufficient to compute the anomaly equations and on the new anomaly categories being useful constructs; no explicit free parameters are named in the abstract.

free parameters (1)
  • Distance/speed thresholds in anomaly equations
    Required to operationalize A3 close approach and similar definitions but not quantified in abstract.
axioms (1)
  • domain assumption Standard AIS fields (position, timestamp, speed) suffice to implement the three anomaly equations
    Invoked when stating the taxonomy is implementable under a limited observation schema.
invented entities (1)
  • Anomaly taxonomy (A1, A2, A3) no independent evidence
    purpose: To provide objective, equation-computable categories for single- and inter-vessel anomalies
    Newly introduced constructs; abstract provides no external validation or falsifiable prediction for them.

pith-pipeline@v0.9.1-grok · 5824 in / 1349 out tokens · 59327 ms · 2026-06-30T07:48:47.932783+00:00 · methodology

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

60 extracted references · 8 canonical work pages · 2 internal anchors

  1. [1]

    Julien Audibert, Pietro Michiardi, Frédéric Guyard, Sébastien Marti, and Maria A Zuluaga. 2020. Usad: Unsupervised anomaly detection on multivariate time series. InProceedings of the 26th ACM SIGKDD international conference on knowledge discovery & data mining. 3395–3404

    2020

  2. [2]

    Zhen Bi, Ningyu Zhang, Yida Xue, Yixin Ou, Daxiong Ji, Guozhou Zheng, and Huajun Chen. 2024. Oceangpt: A large language model for ocean science tasks. InProceedings of the 62nd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers). 3357–3372

    2024

  3. [3]

    Ane Blázquez-García, Angel Conde, Usue Mori, and Jose A Lozano. 2021. A review on outlier/anomaly detection in time series data.ACM computing surveys (CSUR)54, 3 (2021), 1–33

    2021

  4. [4]

    Markus M Breunig, Hans-Peter Kriegel, Raymond T Ng, and Jörg Sander. 2000. LOF: identifying density-based local outliers. InProceedings of the 2000 ACM SIGMOD International Conference on Management of Data. 93–104

    2000

  5. [5]

    Nanyu Chen, Anran Yang, Luo Chen, Hui Wu, and Ning Jing. 2025. MSCE: Empowering Vessel Identity Anomaly Detection with Multimodal LLMs.IEEE Trans. Aerospace Electron. Systems(2025)

    2025

  6. [6]

    TG Coldwell. 1983. Marine Traffic Behaviour in Restricted Waters.Journal of Navigation36, 3 (1983), 430–444

    1983

  7. [7]

    Yanai Elazar, Nora Kassner, Shauli Ravfogel, Abhilasha Ravichander, Eduard Hovy, Hinrich Schütze, and Yoav Goldberg. 2021. Measuring and Improving Consistency in Pretrained Language Models.Transactions of the Association for Computational Linguistics9 (2021), 1012–1031. doi:10.1162/tacl_a_00410

    work page doi:10.1162/tacl_a_00410 2021

  8. [8]

    Elisabeth M Goodwin. 1975. A Statistical Study of Ship Domains.Journal of Navigation28, 3 (1975), 328–344

    1975

  9. [9]

    Jisang Ha, Myung-Il Roh, and Hye-Won Lee. 2021. Quantitative calculation method of the collision risk for collision avoidance in ship navigation using the CPA and ship domain.Journal of Computational Design and Engineering8, 3 (2021), 894–909

    2021

  10. [10]

    Alex Havrilla, Andrew Dai, Laura O’Mahony, Koen Oostermeijer, Vera Zisler, Alon Albalak, Fabrizio Milo, Sharath Chandra Raparthy, Kanishk Gandhi, Baber Abbasi, et al. 2024. Surveying the effects of quality, diversity, and complexity in synthetic data from large language models.arXiv preprint arXiv:2412.02980 (2024)

    work page arXiv 2024

  11. [11]

    Sepp Hochreiter and Jürgen Schmidhuber. 1997. Long short-term memory.Neural computation9, 8 (1997), 1735–1780

    1997

  12. [12]

    International Maritime Organization. 1972. Convention on the International Regulations for Preventing Collisions at Sea, 1972 (COLREGs). https://www.imo. org/en/about/conventions/pages/colreg.aspx Accessed: 2026-01-07

    1972

  13. [13]

    International Maritime Organization. 1995. Resolution A.823(19): Perfor- mance Standards for Automatic Radar Plotting Aids (ARPAs). IMO Assem- bly Resolution. https://wwwcdn.imo.org/localresources/en/KnowledgeCentre/ IndexofIMOResolutions/AssemblyDocuments/A.823(19).pdf Adopted 23 Novem- ber 1995

    1995

  14. [14]

    Jeehong Kim, Youngseok Hwang, Minchan Kim, Sungho Bae, and Hyunwoo Park

  15. [15]

    Spatio-Temporal Graphs Beyond Grids: Benchmark for Maritime Anomaly Detection.arXiv preprint arXiv:2512.20086(2025)

    work page arXiv 2025

  16. [16]

    Jeehong Kim, Minchan Kim, Youngseok Hwang, Sungho Bae, Deuk Jae Cho, Wonhee Lee, and Hyunwoo Park. 2026. STGVAD: Spatio-Temporal Graph-Based Vessel Behavior Anomaly Detection.IEEE Access14 (2026), 2152–2165. doi:10. 1109/ACCESS.2025.3609783

    work page arXiv 2026

  17. [17]

    Jinhee Kim, Taesung Kim, and Jaegul Choo. 2024. Epic: Effective prompting for imbalanced-class data synthesis in tabular data classification via large language models.Advances in Neural Information Processing Systems37 (2024), 31504– 31542

    2024

  18. [18]

    Richard O Lane, David A Nevell, Steven D Hayward, and Thomas W Beaney. 2010. Maritime anomaly detection and threat assessment. In2010 13th International conference on information fusion. IEEE, 1–8

    2010

  19. [19]

    Hui Li, Wengen Li, Shuyu Wang, Hanchen Yang, Jihong Guan, and Yichao Zhang

  20. [20]

    STAD: Ship trajectory anomaly detection in ocean with dynamic pattern clustering.Ocean Engineering313 (2024), 119530

    2024

  21. [21]

    Maohan Liang, Lingxuan Weng, Ruobin Gao, Yan Li, and Liang Du. 2024. Unsu- pervised maritime anomaly detection for intelligent situational awareness using AIS data.Knowledge-Based Systems284 (2024), 111313

    2024

  22. [22]

    Fei Tony Liu, Kai Ming Ting, and Zhi-Hua Zhou. 2008. Isolation Forest. In2008 Eighth IEEE International Conference on Data Mining. 413–422. doi:10.1109/ICDM. 2008.17

    work page doi:10.1109/icdm 2008

  23. [23]

    Jinliang Liu, Jianghui Li, and Chunshan Liu. 2024. AIS-based kinematic anomaly classification for maritime surveillance.Ocean Engineering305 (2024), 118026

    2024

  24. [24]

    Shizhan Liu, Hang Yu, Cong Liao, Jianguo Li, Weiyao Lin, Alex X Liu, and Schahram Dustdar. 2022. Pyraformer: Low-complexity pyramidal attention for long-range time series modeling and forecasting. InInternational Conference on Learning Representations

    2022

  25. [25]

    Lin Long, Rui Wang, Ruixuan Xiao, Junbo Zhao, Xiao Ding, Gang Chen, and Haobo Wang. 2024. On LLMs-Driven Synthetic Data Generation, Curation, and Evaluation: A Survey. InFindings of the Association for Computational Linguistics: ACL 2024. Association for Computational Linguistics, 11065–11082

    2024

  26. [26]

    Quandang Ma, Sunrong Lian, Dingze Zhang, Xiao Lang, Hao Rong, Wengang Mao, and Mingyang Zhang. 2025. A machine learning method for the recognition of ship behavior using AIS data.Ocean Engineering315 (2025), 119791. doi:10. 1016/j.oceaneng.2024.119791

    work page arXiv 2025

  27. [27]

    Martin Masek, Chiou Peng Lam, Travis Rybicki, Jacob Snell, Daniel Wheat, Luke Kelly, and Cheryl Smith-Gander. 2021. The Open Maritime Traffic Analysis Dataset. InProceedings of the 24th International Congress on Modelling and Simu- lation (MODSIM2021). Sydney, NSW, Australia

    2021

  28. [28]

    Lucas May Petry, Amilcar Soares, Vania Bogorny, Bruno Brandoli, and Stan Matwin. 2020. Challenges in vessel behavior and anomaly detection: From classical machine learning to deep learning. InCanadian Conference on Artificial Intelligence. Springer, 401–407

    2020

  29. [29]

    Yuqi Nie, Nam H Nguyen, Phanwadee Sinthong, and Jayant Kalagnanam. 2023. A Time Series is Worth 64 Words: Long-term Forecasting with Transformers. In The Eleventh International Conference on Learning Representations

    2023

  30. [30]

    Daehyung Park, Yuuna Hoshi, and Charles C Kemp. 2018. A multimodal anomaly detector for robot-assisted feeding using an lstm-based variational autoencoder. IEEE Robotics and Automation Letters3, 3 (2018), 1544–1551

    2018

  31. [31]

    Yuhao Qi, Jiaxuan Yang, Dongsheng Xu, Ran Shao, Yangyang Duan, and Yangjie Wang. 2026. Vessel Trajectory Anomaly Detection Based on Multi-scale Convo- lutional Autoencoder.Ocean Engineering343 (2026), 123564

    2026

  32. [32]

    Claudio V Ribeiro, Aline Paes, and Daniel de Oliveira. 2023. AIS-based maritime anomaly traffic detection: A review.Expert Systems with Applications231 (2023), 120561

    2023

  33. [33]

    Maria Riveiro, Giuliana Pallotta, and Michele Vespe. 2018. Maritime anomaly detection: A review.Wiley Interdisciplinary Reviews: Data Mining and Knowledge Discovery8, 5 (2018), e1266

    2018

  34. [34]

    Cécile Rousseau, Tobia Boschi, Giandomenico Cornacchia, Dhaval Salwala, Alessandra Pascale, and Juan Bernabe Moreno. 2025. Forging Time Series with Language: A Large Language Model Approach to Synthetic Data Generation. In The Thirty-ninth Annual Conference on Neural Information Processing Systems

    2025

  35. [35]

    Bernhard Schölkopf, John C Platt, John Shawe-Taylor, Alex J Smola, and Robert C Williamson. 2001. Estimating the support of a high-dimensional distribution. Neural Computation13, 7 (2001), 1443–1471

    2001

  36. [36]

    Abdoulaye Sidibé and Gao Shu. 2017. Study of automatic anomalous behaviour detection techniques for maritime vessels.The journal of Navigation70, 4 (2017), 847–858

    2017

  37. [37]

    Ya Su, Youjian Zhao, Chenhao Niu, Rong Liu, Wei Sun, and Dan Pei. 2019. Robust anomaly detection for multivariate time series through stochastic recurrent neural network. InProceedings of the 25th ACM SIGKDD International Conference on Knowledge Discovery & Data Mining. 2828–2837

    2019

  38. [38]

    Bowen Sui, Jianqiang Zhang, and Zhong Liu. 2023. A real-time ship encounter collision risk detection approach in close-quarters situation.Measurement and Control56, 9-10 (2023), 1613–1625

    2023

  39. [39]

    Weiwei Tian, Beatriz Sanguino, Mingda Zhu, Øivind Kåre Kjerstad, Guoyuan Li, and Houxiang Zhang. 2025. Knowledge extraction from decision-making data for maritime navigation support.Ocean Engineering331 (2025), 121268

    2025

  40. [40]

    Shreshth Tuli, Giuliano Casale, and Nicholas R Jennings. 2022. TranAD: deep transformer networks for anomaly detection in multivariate time series data. Proceedings of the VLDB Endowment15, 6 (2022), 1201–1214

    2022

  41. [41]

    Ashish Vaswani, Noam Shazeer, Niki Parmar, Jakob Uszkoreit, Llion Jones, Aidan N Gomez, Łukasz Kaiser, and Illia Polosukhin. 2017. Attention is all you need.Advances in neural information processing systems30 (2017)

    2017

  42. [42]

    Petar Veličković, Guillem Cucurull, Arantxa Casanova, Adriana Romero, Pietro Lio, and Yoshua Bengio. 2017. Graph attention networks.arXiv preprint arXiv:1710.10903(2017)

    work page internal anchor Pith review Pith/arXiv arXiv 2017

  43. [43]

    Yuanqiao Wen, Wei Tao, Zhongyi Sui, Miquel Angel Piera, and Rongxin Song

  44. [44]

    Dynamic model-based method for the analysis of ship behavior in marine traffic situation.Ocean Engineering257 (2022), 111578

    2022

  45. [45]

    Haixu Wu, Jiehui Xu, Jianmin Wang, and Mingsheng Long. 2021. Autoformer: De- composition transformers with auto-correlation for long-term series forecasting. Advances in Neural Information Processing Systems34 (2021), 22419–22430

    2021

  46. [46]

    Zhexin Xie, Xiangen Bai, Xiaofeng Xu, and Yingjie Xiao. 2024. An anomaly detection method based on ship behavior trajectory.Ocean Engineering293 (2024), 116640

    2024

  47. [47]

    Dongsheng Xu, Jiaxuan Yang, Ken Sinkou Qin, Yuhao Qi, and Ziyao Zhou. 2025. Anomaly detection method for ship trajectory based on stay region mining. Ocean Engineering331 (2025), 121364

    2025

  48. [48]

    Jiehui Xu, Haixu Wu, Jianmin Wang, and Mingsheng Long. 2022. Anomaly Transformer: Time Series Anomaly Detection with Association Discrepancy. In International Conference on Learning Representations

    2022

  49. [49]

    An Yang, Anfeng Li, Baosong Yang, Beichen Zhang, Binyuan Hui, Bo Zheng, Bowen Yu, Chang Gao, Chengen Huang, Chenxu Lv, et al. 2025. Qwen3 technical report.arXiv preprint arXiv:2505.09388(2025)

    work page internal anchor Pith review Pith/arXiv arXiv 2025

  50. [50]

    June Yong Yang, Geondo Park, Joowon Kim, Hyeongwon Jang, and Eunho Yang

  51. [51]

    Language-interfaced tabular oversampling via progressive imputation and self-authentication. InThe Twelfth International Conference on Learning Redefining Maritime Anomaly Detection via Equation-Grounded Synthetic Anomalies KDD ’26, August 09–13, 2026, Jeju Island, Republic of Korea Representations

    2026

  52. [52]

    Ying Yang, Yang Liu, Guorong Li, Zekun Zhang, and Yanbin Liu. 2024. Harness- ing the power of Machine learning for AIS Data-Driven maritime Research: A comprehensive review.Transportation research part E: logistics and transportation review183 (2024), 103426

    2024

  53. [53]

    Sang-Lok Yoo. 2018. Near-miss density map for safe navigation of ships.Ocean Engineering163 (2018), 15–21

    2018

  54. [54]

    Qiaochan Yu, Xiangjun Yin, Xiongfei Geng, Siyuan Chen, and Jingyu Yang. 2025. AISFormer for long-term vessel trajectory prediction.Ocean Engineering340 (2025), 122098

    2025

  55. [55]

    Ailing Zeng, Muxi Chen, Lei Zhang, and Qiang Xu. 2023. Are transformers effective for time series forecasting?. InProceedings of the AAAI conference on artificial intelligence, Vol. 37. 11121–11128

    2023

  56. [56]

    Tianyi Zeng and Yao Zhang. 2026. Large language model-augmented model predictive control for marine vessels in uncertain marine environments.Ocean Engineering346 (2026), 123628

    2026

  57. [57]

    Weibin Zhang, Floris Goerlandt, Jakub Montewka, and Pentti Kujala. 2015. A method for detecting possible near miss ship collisions from AIS data.Ocean Engineering107 (2015), 60–69

    2015

  58. [58]

    Haoyi Zhou, Shanghang Zhang, Jieqi Peng, Shuai Zhang, Jianxin Li, Hui Xiong, and Wancai Zhang. 2021. Informer: Beyond efficient transformer for long se- quence time-series forecasting. InProceedings of the AAAI Conference on Artificial Intelligence, Vol. 35. 11106–11115

    2021

  59. [59]

    Tian Zhou, Ziqing Ma, Qingsong Wen, Xue Wang, Liang Sun, and Rong Jin

  60. [60]

    Anomaly Timestamp Scoring Module

    FEDformer: Frequency enhanced decomposed transformer for long-term series forecasting. InInternational Conference on Machine Learning. PMLR, 27268– 27286. A Additional Details of Baselines • LOF[ 4]: Density-based detector that scores each point by comparing its local reachability density to neighboring points. • OCSVM[ 33]: Learns a decision boundary enc...

    2026