You do not have permission to edit this page, for the following reason:
The action you have requested is limited to users in the group: Users.
Project description (free text)
Give a concise project description. Include:
NOTE: please note that this thesis *requires* a strong prior knowledge of ML and data mining techniques (especially for anomaly detection). In addition, good programming skills (e.g. in Python) are also required. Contacts: * mohamed-rafik.bouguelia@hh.se * onur.dikmen@hh.se Description: Anomaly detection allows to find patterns that deviate significantly from the majority of reference data, indicating e.g. a system fault. Conventional anomaly detection methods focus on statistical features of the data; they are unsupervised (due to the expensive labeling costs of ground truth anomalies) and do not interact with a human-expert. However, from the user perspective, a detected anomaly can either be relevant (i.e. an actual anomaly) or irrelevant (e.g. an atypical but normal system behavior). In interactive anomaly detection, the goal is to maximize the true (relevant) anomalies presented to the human expert (user). This is done by allowing the algorithm to proactively communicate with the user and leverage the user-feedback to refine results and learn to suggest more relevant anomalies. Several challenges comes with this. For example, one want to learn to distinguish relevant from irrelevant anomalies, but without presenting a lot of irrelevant anomalies to the user. This raises the question of how to handle the exploration-exploitation dilemma when querying anomalies of different kinds. The goal of this thesis is to: # Perform a state-of-the-art literature review of existing methods for interactive anomaly detection. # Identify the advantages and limitations of these existing methods. # Propose a new interactive anomaly detection method wich solves some of the identified limitations. # Perform extensive experiments on artificial and real-world datasets to show the advantage of the proposed method over existing methods for interactive anomaly detection. Some relevant recent (2019) papers: * Lamba, H. and Akoglu, L., 2019, May. Learning On-the-Job to Re-rank Anomalies from Top-1 Feedback. In Proceedings of the 2019 SIAM International Conference on Data Mining (SDM), pp. 612-620. Society for Industrial and Applied Mathematics. [https://epubs.siam.org/doi/pdf/10.1137/1.9781611975673.69 https://epubs.siam.org/doi/pdf/10.1137/1.9781611975673.69] * Ding, K., Li, J. and Liu, H., 2019, January. Interactive anomaly detection on attributed networks. In Proceedings of the Twelfth ACM International Conference on Web Search and Data Mining, pp. 357-365. ACM. [http://www.public.asu.edu/~jundongl/paper/WSDM19_GraphUCB.pdf http://www.public.asu.edu/~jundongl/paper/WSDM19_GraphUCB.pdf] * Arnaldo, I., Veeramachaneni, K. and Lam, M., 2019. ex2: a framework for interactive anomaly detection. In ACM IUI Workshop on Exploratory Search and Interactive Data Analytics (ESIDA). [http://ceur-ws.org/Vol-2327/IUI19WS-ESIDA-2.pdf http://ceur-ws.org/Vol-2327/IUI19WS-ESIDA-2.pdf] * Zhu, Y. and Yang, K., 2019. Tripartite Active Learning for Interactive Anomaly Discovery. IEEE Access. [https://ieeexplore.ieee.org/stamp/stamp.jsp?arnumber=8707963 https://ieeexplore.ieee.org/stamp/stamp.jsp?arnumber=8707963]
Summary:
This is a minor edit Watch this page
Cancel
Home
Research
Education
Partners
People
Contact
