CAREER: A General Framework for Methodical and Interpretable Anomaly Mining
is based upon work supported by the National Science Foundation
under Grant No. 1452425. Any opinions, findings, and
conclusions or recommendations expressed in this material are those
of the author(s) and do not necessarily reflect the views of the
National Science Foundation.
Anomaly Mining, Novel Anomaly Definitions, Descriptive Methods, Anomaly Ensembles, Anomaly Mining Framework, Heterogeneous Data Mining,
Graph Mining, Applications.
- NSF, Award Number: CAREER 1452425, Duration:
In addition to the PIs, the following graduate students work on
- Hung Nguyen (CMU, Post doc, 2018-2019)
- Tuan Le (CMU, Post doc, 2017-2018)
- Meghanath Macha (CMU, PhD student, 2016-2018)
- Emaad Ahmed Manzoor (CMU, PhD student, 2015-2017)
- Shebuti Rayana (SBU, PhD, 2012-2016)
- Junting Ye (SBU, PhD, 2014-2019)
This proposal aims to push the boundaries of anomaly mining as a field through a quest for principled foundations and practices. Research will create previously unstudied classes of data representations that unify heterogeneous data sources, and build on them to formulate novel anomaly mining problems. We will invent new, descriptive algorithms for complex anomaly detection and characterization, that will also explore and exploit ensemble and multi-view approaches. The proposed research will give rise to a comprehensive framework for anomaly mining; through a deeper understanding of the space of
problems and objectives, new models and algorithms, and systematic techniques to harness them.
: Proposed research will take the essential steps to mature anomaly mining into a valuable contributor to the larger world. It will have direct significance to many concrete problems (e.g., outsider threat, fraud, intrusion) important for the government, industry, and the society. The project will build a web-based platform that hosts a repository of formulations, algorithms, tools, and datasets, for the research community and the public to leverage. We will collaborate with industry and hospital partners to shepherd our innovations into deployed technology, with tangible impact on security and healthcare.
: The PI is committed to developing an education plan that: trains students to think creatively in formulating and solving problems, enhances undergraduate training by involving Honors thesis students in proposed research, promotes campus wide synergism for students across departments, and increases the role of women in Computer Science through mentoring and open house events for women in the community.
The project summary can be found here
- 2016 Workshop ODD 4.0 on outlier definition, detection and description on demand
co-organized in conjunction with SIGKDD 2016
- 2015 Workshop ODDx3 on outlier definition, detection and description
co-organized in conjunction with SIGKDD 2015
- 2014 Workshop ODD^2 on outlier detection and description under data diversity
co-organized in conjunction with SIGKDD 2014
- 2013 Workshop ODD on outlier detection and description
co-organized in conjunction with SIGKDD 2013
The educational contributions of the project include:
- PI Akoglu is teaching a course titled 95-828 Machine Learning for Problem Solving
at CMU's Heinz College, which integrates outlier detection in its curriculum.
- Doctoral dissertations:
- Junting Ye, Network Analysis and Modeling for News and Social Media, April 2019.
- Shebuti Rayana, Ensemble and Multimodal Learning for Anomaly Mining: Algorithms and Applications. August, 2017.
Point of Contact: Leman Akoglu, lakoglu AT cs.cmu.edu
Last updated: May 2020, by Leman Akoglu