HIIT Special Seminar: Mohammad Sabokrou "Reliable Deep Anomaly Detection: Solutions and Future Challenges"
Reliable Deep Anomaly Detection: Solutions and Future Challenges
Monday 2 May 13:00-14:00
via Zoom: request the link by email: [email protected]
Note! The link will be sent by email to Aalto CS, Helsinki CS, Aalto ComNet, and Aalto SPA staff
Abstract: Deep neural networks have achieved great performance where training and testing samples are sampled from the same distribution. Nevertheless, sometimes models encounter diverged samples from the training distribution. Failure to recognize an abnormal sample (i.e, out-of-training samples distribution), and consequently assign that sample to an in-class label significantly compromises the reliability of a model. Besides, anomaly detection is a fundamental machine learning task that has a great effect on downstream tasks. The problem has gained significant attention due to its importance for safety deploying models in open-world settings. In this talk, I will introduce out of distribution detection, novelty detection, and open set recognition all under the umbrella of anomaly detection. Then I will explore the current challenges and finally shed light on future lines of research.
Bio: Mohammad Sabokrou completed his PhD in Artificial Intelligence in 2017 from the Malek-Ashtar University of Technology in Tehran. Since graduating, he was a Postdoctoral Research at the Institute for Research in Fundamental Science (IMP) in Tehran. He has also been a visiting researcher at the University of Oulu in Finland and the University of Technology of Troyes in France.
His principal area of research is machine learning and computer vision topics such as anomaly detection, video analysing, self-supervised learning, etc.