Serdecznie zapraszamy na seminarium Wydziału Informatyki AGH, które odbędzie się 23 czerwca 2025 r. (tj. poniedziałek) od godz. 13:30 do 14:30 w sali 2.41.
Gościem najbliższego spotkania będzie dr Roberto Corizzo z American University w Waszyngtonie, który przedstawi seminarium o tytule “Continual Lifelong Learning for Anomaly Detection”.
Tematyka seminarium:
Anomaly detection is critical in dynamic and evolving environments such as cyber-physical systems, human monitoring, and network traffic analysis. While traditional anomaly detection approaches rely on offline learning with static data or online learning with continuous adaptation, lifelong and continual learning are emerging as promising paradigms to develop models that can adapt to new tasks and environments while preserving previously acquired knowledge. Despite their potential, these paradigms have been primarily explored in image classification and reinforcement learning, leaving their application to anomaly detection underexplored.
This seminar will introduce continual anomaly detection as a key approach to building robust, adaptable models capable of detecting anomalous behavior across changing conditions. The seminar will begin by motivating the need for continual anomaly detection, characterizing its tasks, and highlighting real-world applications in cybersecurity, smart grids, and cyber-physical systems.
We will then explore the unique challenges of continual anomaly detection, such as class imbalance due to scarce anomalies, lack of task identity information (task-incremental vs. task-agnostic settings), and the evolving definition of the normal class across tasks. These aspects challenge models to maintain stability while adapting to new normal behaviors and unforeseen anomalies.
Next, we will present practical tools for tackling these challenges, including a scenario extraction procedure that enables researchers to generate continual learning settings from existing datasets. This will be complemented by a demonstration of a semi-supervised continual anomaly detection workflow and a discussion of suitable evaluation metrics for such scenarios.
Finally, we will introduce pyCLAD, an open-source software framework designed to facilitate continual lifelong anomaly detection experimental pipelines.
Throughout the seminar, relevant research works will be discussed in an intuitive manner, providing both theoretical insight and practical guidance. Open problems and future research directions will be highlighted to encourage further exploration and open research opportunities in this emerging field.
BIO prelegenta:
Roberto Corizzo,Ph.D., is an Assistant Professor, Dept. of Computer Science, American University. Prior to that, he was a Research Fellow in the Department of Computer Science at University of Bari, Italy.
He conducts research on machine learning, data mining, big data analytics, and continual learning. His research addresses analytical tasks such as forecasting and anomaly detection tailored to real-world applications in fields such as energy, cybersecurity, astrophysics, and social networks.
He co-authored 75 articles, including 28 publications in journals such as IEEE Transactions on Neural Networks and Learning Systems, Neural Networks, and Machine Learning. He participated in the scientific committee of international conferences and served as a reviewer for several international journals.