Challenges in Vessel Behavior and Anomaly Detection: From Classical - PowerPoint PPT Presentation

Challenges in Vessel Behavior and Anomaly Detection: From Classical Machine Learning to Deep Learning Lucas May Petry 1 , Amilcar Soares 2 , Vania Bogorny 1 , Bruno Brandoli 2 , Stan Matwin 2 1 Programa de Pós-Graduação em Ciências da Computação (PPGCC), Universidade Federal de Santa Catarina (UFSC), Florianópolis, Brazil 2 Institute for Big Data Analytics, Dalhousie University, Halifax, Canada 33rd Canadian Conference on Artificial Intelligence CAI 2020 | Petry et al. | Challenges in Vessel Behavior and Anomaly Detection

Outline ● Introduction ● Related Work ● Objective and Contributions ● Research Challenges ● Conclusion CAI 2020 | Petry et al. | Challenges in Vessel Behavior and Anomaly Detection 2

Introduction ● Maritime transportation represents 90% of all international trade volume [22] ● Expansion of maritime activities and development of the Automatic Identification System (AIS) ● Development of maritime monitoring systems Prevent vessel accidents ○ Detect illegal activities ○ Protect the marine fauna and flora ○ ● High volume of data makes real-time monitoring more challenging to maritime agents Detect anomalies, changes of behavior, events ○ CAI 2020 | Petry et al. | Challenges in Vessel Behavior and Anomaly Detection 3

Introduction CAI 2020 | Petry et al. | Challenges in Vessel Behavior and Anomaly Detection 4

Related Work No data Multi-sensor Work Main approach Unsupervised pretraining data Terroso et al. [24] Algorithm/rules Yes Yes* Patroumpas et al. [16] Queries/rules Yes Varlamis et al. [25]** Algorithm/rules Yes Yes Event detection Wen et al. [19] Probability model Yes Lei [9]** Clustering Yes Yes Soares et al. [18] Queries/rules Yes Yes Bomberger et al. [28] Neural network Yes Kernel Density Ristic et al. [26] Yes Estimation Anomaly detection Riveiro et al. [27] Gaussian model Nguyen et al. [14] Neural network Yes Varlamis et al. [29] Clustering Yes *They only use the weather description (e.g. sunny, rainy) as a deciding factor about detected abnormal low speed behavior. **Only a single type of behavior/event is detected. CAI 2020 | Petry et al. | Challenges in Vessel Behavior and Anomaly Detection 5

Objective and Contributions Objective Present research gaps and challenges in machine learning for detecting different types of vessel behavior, considering several constraints imposed by real-time data streams and the maritime monitoring domain. Contributions ● Short survey of the state of the art ● Extensive discussion on major topics with opportunities of research on vessel behavior detection with machine learning CAI 2020 | Petry et al. | Challenges in Vessel Behavior and Anomaly Detection 6

Research Challenges Tasks Data Issues Limited or non-existent Detection of vessel behaviors labeled data Absent knowledge of Detection of behavior behaviors or labels present in recurrence the data Provide means for Highly-dimensional data interpreting detected points derived from multiple behaviors sources CAI 2020 | Petry et al. | Challenges in Vessel Behavior and Anomaly Detection 7

Research Challenges ● Behavior detection Concept drift techniques [1, 8, 17] ○ Limited to univariate data, lack of interpretability ■ They can only detect points of change ■ ● Recurrent behaviors Toeplitz Inverse Covariance-based Clustering (TICC) [6] ○ Markov Random Fields (MRFs) for representing clusters/behaviors ■ MRFs may provide insight for behavior interpretation! ● Number of behaviors should be known apriori ■ Assumes all data is available ■ CAI 2020 | Petry et al. | Challenges in Vessel Behavior and Anomaly Detection 8

Research Challenges ● Deep learning No work has used it for vessel behavior detection ○ Convolutional Neural Networks (CNNs) ○ Satellite imagery can be expensive and make detecting certain ■ behaviors very difficult or even impossible [34] They have been used for trajectory classification/prediction based ■ on movement features [3, 30] Visual techniques can be used for achieving interpretability [21, 31] ■ ● Big, yet limited data High volume of data available, but lacks labels ○ Use knowledge of previous works as a ground truth ○ Transfer learning for learning from a few examples [32, 33] ○ Generative Adversarial Networks (GANs) for synthesizing new ○ behavior data CAI 2020 | Petry et al. | Challenges in Vessel Behavior and Anomaly Detection 9

Conclusion ● Maritime monitoring has experienced significant progress in the last decade ● Existing works do not take full advantage of machine learning techniques for vessel behavior detection ● We presented several research gaps in the field, indicating opportunities for future works ● We hope to instigate the development of new algorithms, methods, and tools for maritime monitoring CAI 2020 | Petry et al. | Challenges in Vessel Behavior and Anomaly Detection 10

References 1. Bifet, A., Gavalda, R.: Learning from time-changing data with adaptive windowing. In: Proc. of the 2007 SIAM int. conf. on data mining. pp. 443–448. SIAM (2007) 2. Claramunt, C., Ray, C., Salmon, L., Camossi, E., Hadzagic, M., Jousselme, A., Andrienko, G., Andrienko, N., Theodoridis, Y., Vouros, G.: Maritime data integration and analysis: recent progress and research challenges. Adv. in Database Technology - EDBT 2017, 192–197 (2017) 3. Dabiri, S., Heaslip, K.: Inferring transportation modes from gps trajectories using a convolutional neural network. Transportation research part C: emerging technologies 86, 360–371 (2018) 4. Doshi-Velez, F., Kim, B.: Towards a rigorous science of interpretable machine learning. arXiv preprint arXiv:1702.08608 (2017) 5. Goodfellow, I., Pouget-Abadie, J., Mirza, M., Xu, B., Warde-Farley, D., Ozair, S., Courville, A., Bengio, Y.: Generative adversarial nets. In: Adv. in neural inf. processing systems. pp. 2672–2680 (2014) 6. Hallac, D., Vare, S., Boyd, S., Leskovec, J.: Toeplitz inverse covariance-based clustering of multivariate time series data. In: Proc. of the 23rd ACM SIGKDD Int. Conf. on Knowledge Discovery and Data Mining. pp. 215–223. ACM (2017) 7. Krizhevsky, A., Sutskever, I., Hinton, G.E.: Imagenet classification with deep convolutional neural networks. In: Adv. in neural inf. processing systems. pp. 1097–1105 (2012) 8. Lee, J., Magoules, F.: Detection of concept drift for learning from stream data. In: 14th Int. Conf. on High Performance Computing and Communication & 9th Int. Conf. on Embedded Software and Systems. pp. 241–245. IEEE (2012) CAI 2020 | Petry et al. | Challenges in Vessel Behavior and Anomaly Detection 11

References 9. Lei, P.R.: Mining maritime traffic conflict trajectories from a massive ais data. Knowledge and Inf. Systems pp. 1–27 (2019) 10. Lipton, Z.C.: The mythos of model interpretability. arXiv preprint arXiv:1606.03490 (2016) 11. Lv, J., Li, Q., Sun, Q., Wang, X.: T-conv: a convolutional neural network for multi-scale taxi trajectory prediction. In: 2018 IEEE int. conf. on big data and smart computing (bigcomp). pp. 82–89. IEEE (2018) 12. Manzoor, E., Lamba, H., Akoglu, L.: xstream: Outlier detection in feature-evolving data streams. In: Proc. of the 24th ACM SIGKDD Int. Conf. on Knowledge Discovery & Data Mining. pp. 1963–1972. KDD ’18, ACM (2018) 13. Mohammadi, M., Al-Fuqaha, A.: Enabling cognitive smart cities using big data and machine learning: Approaches and challenges. IEEE Communications Magazine 56(2), 94–101 (2018) 14. Nguyen, D., Vadaine, R., Hajduch, G., Garello, R., Fablet, R.: A multi-task deep learning architecture for maritime surveillance using ais data streams. In: 5th Int. Conf. on Data Science and Advanced Analytics (DSAA). pp. 331–340. IEEE (2018) 15. Nyman, E.: Techno-optimism and ocean governance: New trends in maritime monitoring. Marine Policy 99, 30–33 (2019) 16. Patroumpas, K., Artikis, A., Katzouris, N., Vodas, M., Theodoridis, Y., Pelekis, N.: Event recognition for maritime surveillance. In: Adv. in Database Technology - EDBT. pp. 629–640 (2015) CAI 2020 | Petry et al. | Challenges in Vessel Behavior and Anomaly Detection 12