Speaker Details
[more info about the speakers will be announced here]
Chen Qiu
Bosch Research
Bio:
Dr. Chen Qiu is a research scientist at Bosch Research, Pittsburgh, PA, USA since November 2022. He obtained his Ph.D. with summa cum laude in computer science at the University of Kaiserslautern-Landau (Germany) under the supervision of Prof. Marius Kloft. He received his M.S. degree from the University of Erlangen-Nuremberg (Germany) in 2018 and his B.E. degree from Yanshan University (China) in 2014. His research mainly focuses on machine learning and deep learning, with a special emphasis on anomaly detection. In addition to his thesis research on anomalies, his general research interests span various domains including vision-language models and prompt learning.
Keynote Title:
Navigating Data and Label Scarcity in Anomaly Detection.
Keynote Abstract:
Obtaining labeled data for anomaly detection is often challenging, costly, or impractical, yet it is crucial for training accurate models. This presentation explores innovative strategies to overcome data and label scarcity in anomaly detection. Our research leverages latent variable models to improve the performance of anomaly detectors with contaminated training data and introduces a data-agnostic method for making small anomaly detection models effective in zero-shot scenarios. Additionally, we investigate the use of synthetic anomalies to further enhance detection capabilities. These strategies provide robust solutions for developing effective anomaly detection systems despite the limitations posed by scarce labeled data.
Jong-Seok Lee
Yonsei University
Bio:
Prof. Jong-Seok Lee is a professor at the School of Integrated Technology, Yonsei University, Korea. He received a PhD degree in electrical engineering from KAIST, Korea. He was a postdoctoral researcher at EPFL, Switzerland. He served as an elected member of the Multimedia Signal Processing Technical Committee of the IEEE Signal Processing Society. He has authored or coauthored more than 200 publications. His recent research interests include learning-based image/video processing, generative models, adversarial machine learning, efficient deep learning, etc.
Keynote Title:
Evaluation of image generation models and generated images.
Keynote Abstract:
Recently, image generation models have received great attention, including generative adversarial networks, variational auto-encoders, and diffusion models. Accordingly, evaluation of such models becomes increasingly important. This talk presents selected recent works on evaluation of image generation models and generated images. Specific topics include probabilistic modeling of embeddings of generated images, generated image embedding using random networks, and evaluation metrics based on analysis of the embedding space.
Guansong Pang
Singapore Management University
Bio:
Dr. Guansong Pang is a tenure-track Assistant Professor of Computer Science at the School of Computing and Information Systems, Singapore Management University (SMU). He is also a faculty member of Centre on Security, Mobile Applications and Cryptography. He was a Research Fellow with the Australian Institute for Machine Learning (AIML), University of Adelaide, Australia. Before joining AIML, he received his Ph.D. at University of Technology Sydney (UTS), Australia. He leads the Machine Learning & Applications (MaLA) Lab at SMU. His research interests include machine learning, data mining and computer vision, with a research theme focused on recognizing and generalizing to abnormal/unknown/unseen data for creating trustworthy AI systems. Some research areas of particular interest include: anomaly detection, open-world learning (out-of-distribution detection/generalization, open-set recognition, long-tailed classification, continual learning, open-vocabulary learning), graph representation learning, deep reinforcement learning for knowledge discovery. He also explores some pivotal real-world applications of these areas, such as network intrusion detection, fraud detection, early detection of diseases/faults, learning from biomedicine data, industrial defect detection, biometric anti-spoofing, hate speech detection, etc. His research has received multiple global recognition/awards, e.g., the prestigious 2020 UTS Chancellor's Award List, the World's Top 2% Scientists in 2022 and 2023, DSAA 2023 Best Paper Award (Applications Track), and Most Influential KDD Papers. He actively engages in various professional activities, serving as Senior PC member/Area Chair of IJCAI'21, PAKDD'23-24, CVPR'24, and NeurIPS'24, and regular PC member/reviewer of premier conferences like ICML, NeurIPS, ICLR, KDD, CVPR, ICCV, ECCV, AAAI, and IJCAI. He is an Associate Editor of IEEE Transactions on Neural Networks and Learning Systems and Pattern Recognition, an Editorial Board member of IEEE Intelligent Systems and International Journal of Data Science and Analytics.
Keynote Title:
Generalist Anomaly Detection.
Keynote Abstract:
Current anomaly detection (AD) methods are based on a one-model-for-one-dataset paradigm, training one detection model on each target dataset individually. Despite remarkable performance on various benchmarks, these methods require the availability of large training data and the skilled detection model training per dataset. Thus, they become infeasible in application scenarios where training on the target dataset is not allowed due to issues like data privacy or unavailability of large training data in the deployment of new applications. They also become ineffective when there is distribution shift in the test data. This talk introduces the paradigm of generalist AD that helps address these issues. The key idea is to build a one-for-all model for AD, i.e., to train one single detection model that can generalize to detect anomalies in diverse datasets from different application domains without any further training/tuning on the target data. Recent methods and empirical results under both zero- and few-shot settings of generalist AD will be introduced.