(Local Time) 08:00am - 12:30pm on Aug. 14, 2022; Location: 209B

08:00am - 11:00am In-person Talks and Presentations

08:00 – 08:10

• Opening Remarks

Dr. Ye Zhu, PC Chair of ANDEA'22

Senior Lecturer, Deakin University

08:10 – 09:10

• Presentation of Accepted Papers (15 minutes per paper, including 5 minutes for Q&A):

• Bin Li, Carsten Jentsch and Emmanuel Müller. Prototypes as Explanation for Time Series Anomaly Detection.

Presented by Bin Li

• Parastoo Kamranfar, David Lattanzi, Amarda Shehu and Daniel Barbara. Multiple Instance Learning for Detecting Anomalies over Sequential Real-World Datasets. [slides]

Presented by Dr. Daniel Barbara

• Ruichuan Zhang, Fangzhou Cheng and Aparna Pandey. Representation Learning Using a Multi-Branch Transformer for Industrial Time Series Anomaly Detection.

Presented by Aparna Pandey

• Hongjing Zhang, Fangzhou Cheng and Aparna Pandey. One-Class Predictive Autoencoder Towards Unsupervised Anomaly Detection on Industrial Time Series. (Workshop Best Paper Award)

Presented by Dr. Hongjing Zhang

09:10 – 09:30

• Test-of-time Award Talk I

Dr. Ye Zhu (on behalf of Dr. Kai Ming Ting)

Fei Tony Liu, Kai Ming Ting, and Zhi-Hua Zhou. Isolation-based Anomaly Detection (TKDD 2012; ICDM 2008) [slides]

09:30 – 10:00

• Coffee Break

10:00 – 10:30

• Keynote 1 - Detecting Anomalies in Complex Systems with Higher Order Networks

Dr. Nitesh Chawla

Frank Freimann Professor, University of Notre Dame

10:30 – 11:00

• Keynote 2 - Acting in Open Worlds in the Face of Disruptive Novelties and Anomalies

Dr. Bryan Loyall

Principal Scientist, Charles River Analytics

11:00am - 12:25pm Pre-recorded/Online Talks and Presentations

11:00 – 11:30

• Keynote 3 - Ensembles for Outlier Evaluation

Dr. Charu Aggarwal

Distinguished Research Staff Member, IBM T. J. Watson Research Center

11:30 – 11:50

• Test-of-time Award Talk II [slides]

Dr. Marius Kloft

Ruff Lukas, Robert Vandermeulen, Nico Goernitz, Lucas Deecke, Shoaib Ahmed Siddiqui, Alexander Binder, Emmanuel Müller, and Marius Kloft. Deep One-Class Classification (ICML 2018)

11:50 – 12:10

• Rising Star Award Talk [slides]

Dr. Lukas Ruff

https://scholar.google.com.sg/citations?user=40QzNXMAAAAJ&hl=en&oi=ao

12:10 -- 12:25

• Presentation of Accepted Papers (15 minutes per paper, including 5 minutes for Q&A):

  • Taylor Dinkins, Sharmodeep Bhattacharyya, Shirshendu Chatterjee, Sabrina Reis and Weng-Keen Wong. Towards Explainable Precision Change point Detection through Linear Decomposition. [slides]

Presented by Taylor Dinkins

Keynotes

Detecting Anomalies in Complex Systems with Higher Order Networks

by Dr. Nitesh Chawla, University of Notre Dame

Abstract In this talk, I will present the higher order network representation of complex systems, and why it is necessary to accurately capture the higher and variable order dependencies among components of a complex system. In this talk, I will present our higher order network algorithm and also show that incorporating higher-order dependencies improves the performance of existing network-based methods for detecting anomalous signals in a complex system.

Bio Nitesh Chawla is the Frank M. Freimann Professor of Computer Science and Engineering at the University of Notre Dame and the Founding Director of the Lucy Family Institute for Data and Society. He is an expert in artificial intelligence, data science, and network science, and is motivated by the question of how technological innovations can advance the common good through interdisciplinary research. He is the recipient of multiple awards for research and teaching innovation including outstanding teacher awards (2007 and 2010), a National Academy of Engineering New Faculty Fellowship, and a number of best paper awards and nominations. He is a Fellow of IEEE. He also is the recipient of the 2015 IEEE CIS Outstanding Early Career Award; the IBM Watson Faculty Award; the IBM Big Data and Analytics Faculty Award; ; and the 1st Source Bank Technology Commercialization Award. He is co-founder of Aunalytics, a data science software and cloud computing company.

Acting in Open Worlds in the Face of Disruptive Novelties and Anomalies

by Dr. Bryan Loyall, Charles River Analytics

Abstract AI Systems today are often brittle to anomalies and novelties that were not previously anticipated by their designers. This brittleness can be a source of both reduced performance and (at times, catastrophic) failure. This brittleness is particularly acute in systems that sense and act autonomously (or semi-autonomously) in open worlds. In this presentation, we outline a range of challenges to creating AI systems that can rapidly and effectively adapt their decision-making and action choices to such unanticipated changes, and we describe research toward a general framework for AI systems that can automatically recognize and adapt to these changes when they arise.

Bio Dr. Bryan Loyall is Director of Technology Innovation and Principal Scientist at Charles River Analytics, where his research interests center on artificial intelligence and machine learning with a focus on: the intersection of symbolic AI and data-driven methods; AI that can interact robustly with open, real-world systems; and integrated human and AI systems. Before joining Charles River, he ran a 60-person AI research group at BAE Systems, Inc, with research across a wide range of AI and computer systems, and was V.P. of Research at RA Capital, founding an AI research group for augmenting human analysis, as well as co-founding two AI-based startup companies. Dr. Loyall received his PhD degree in computer science from Carnegie Mellon.

Ensembles for Outlier Evaluation

by Dr. Charu Aggarwal, IBM T. J. Watson Research Center

Abstract This talk explores the use of ensembles for outlier evaluation. Outlier evaluation is inherently complex because of its unsupervised nature and the unpredictable effects of different parameters. Ensembles are not only effective for improving the performance of outlier detection algorithms but also for evaluating them. An evaluation of several well known algorithms is presented.

Bio Dr. Charu Aggarwal is a Distinguished Research Staff Member (DRSM) at the IBM T. J. Watson Research Center in Yorktown Heights, New York. He completed his Bachelor of Technology in Computer Science from the Indian Institute of Technology at Kanpur in 1993 and his PhD in Operations Research (focus: mathematical optimization) from the Massachusetts Institute of Technology in 1996. He has worked extensively in the field of data mining, with particular interests in data streams, privacy, uncertain data and social network analysis. He is a recipient of an IBM Corporate Award (2003) for his work on bio-terrorist threat detection in data streams, a recipient of the IBM Outstanding Innovation Award (2008) for his scientific contributions to privacy technology, and a recipient of two IBM Outstanding Technical Achievement Awards (2008) for his scientific contributions to high-dimensional and data stream analytics. He has received two best paper awards and an EDBT Test-of-Time Award (2014). He is a recipient of the IEEE ICDM Research Contributions Award (2015) and the ACM SIGKDD Innovation Award (2019), which are the two most prestigious awards for influential research in data mining. He is also a recipient of the W. Wallace McDowell Award , the highest award given by the IEEE Computer Society across the field of computer science. He has served as the general or program co-chair of the IEEE Big Data Conference (2014), the ICDM Conference (2015), the ACM CIKM Conference (2015), and the KDD Conference (2016). He is a fellow of the IEEE (2010), ACM (2013), and the SIAM (2015) for "contributions to knowledge discovery and data mining algorithms."

Test-of-time Award/Rising Star Award Talks

Isolation-based Anomaly Detection and Beyond

by Dr. Kai Ming Ting, Nanjing University (presented by Dr. Ye Zhu)

Abstract Isolation-based methods refer to methods that employ an isolation mechanism to construct isolating partitions in the input space. The first method is called Isolation Forest or iForest, a point anomaly detector, reported in IEEE ICDM 2008 and TKDD 2012. The intuition is that anomalies are rare and different from normal points; thus each anomaly is more amenable to isolation than normal points. A point is said to be isolated if it is contained within an isolating partition that isolates it from the rest of the points in a sample.

The first part of this talk presents the point anomaly detector iForest and its improvements beyond tree structures called aNNE and iNNE which employ Voronoi Diagram and hyperspheres, respectively, as the isolation mechanism. Another important ingredient is data subsampling, which has been largely dismissed or down-played. The second part of the talk describes two notable recent developments of Isolation-based methods beyond point anomaly detection, i.e., Isolation Kernel (IK) and Isolation Distributional Kernel (IDK). They are the key in solving some age-old problems in data mining and machine learning. This includes (i) point-set Kernel Clustering which is the first truly linear time clustering algorithm without the contraints of k-means based clustering since k-means clustering was first studied in the 1950's; and (ii) time series representation and similarity measurements which have been studied in either time or frequency domain in more than one century. IDK provides the first paradigm shift in the treatment of time series.

Bio After receiving his PhD from the University of Sydney, Australia, Kai Ming Ting worked at the University of Waikato (NZ), Deakin University, Monash University and Federation University in Australia. He joined Nanjing University in 2020. Research grants received include those from National Science Foundation of China, US Air Force of Scientific Research (AFOSR/AOARD), Australian Research Council, Toyota InfoTechnology Center and Australian Institute of Sport. He is one of the inventors of Isolation Forest, Isolation Kernel and Isolation Distributional Kernel. Isolation Forest is widely used in industries and academia. Isolation Kernel is a unique similarity measure which is derived from a dataset based on the same/similar isolation mechanism as Isolation Forest, and has no closed-form expression. Isolation Kernel and Isolation Distributonal Kernel are the X-factor that enables many problems to be solved more effectively and efficiently than existing algorithms which rely on Gaussian kernel or Euclidean distance. A brief history of isolation-based methods can be found at https://github.com/IsolationKernel/Codes.

One-class classification is binary classification with one class

by Dr. Marius Kloft

Abstract Previous work found that one-class classification is density level set estimation or binary classification of the normal class vs. a uniform distribution. I present a new view of one-class classification as binary classification with one class. The view leads to a general interpretation of one-class classification and particularly a new formulation of deep one-class classification.

Bio Since 2017 Marius Kloft has been a professor of computer science at TU Kaiserslautern, Germany. Previously, he was an adjunct faculty member of the University of Southern California (09/2018-03/2019), an assistant professor at HU Berlin (2014-2017) and a joint postdoctoral fellow (2012-2014) at the Courant Institute of Mathematical Sciences and Memorial Sloan-Kettering Cancer Center, New York, working with Mehryar Mohri, Corinna Cortes, and Gunnar Rätsch. From 2007-2011, he was a PhD student in the machine learning program of TU Berlin, headed by Klaus-Robert Müller. He was co-advised by Gilles Blanchard and Peter L. Bartlett, whose learning theory group at UC Berkeley he visited from 10/2009 to 10/2010. In 2006, he received a master in mathematics from the University of Marburg with a thesis in algebraic geometry.

Deep One-Class Learning and New Endeavors

by Dr. Lukas Ruff

Abstract In this talk, I will first present main results from my dissertation on Deep One-Class Learning, connecting my body of research over the last years. Afterwards, I will showcase the importance and potential of anomaly detection for cancer research, highlighting examples from digital pathology.

Bio Dr. Lukas Ruff received the joint master’s degree in statistics from the Humboldt University of Berlin, the Berlin Institute of Technology (TU Berlin), and Free University of Berlin in 2017. He received the Ph.D. degree (summa cum laude with Hons.) from TU Berlin in 2021 for his work on deep one-class learning. In May 2021 he joined Aignostics, a Berlin-based digital pathology start-up, as a Senior Machine Learning Scientist. Since March 2022, he became the Lead Machine Learning Researcher at Aignostics, heading the data science research & development team. His research interest include anomaly detection as well as robust and trustworthy machine learning with applications to cancer research.