COE Seminar - Towards Reconciliation of Traditional Approaches and Artificial Intelligence in Computer Vision
COE Department Presents A Seminar on
“Towards Reconciliation of Traditional Approaches and Artificial Intelligence in Computer Vision”
Date: Wednesday, December 25, 2024
Time: 01:00 PM to 02:00 PM
Location: Building 59 Room 1001
Speaker:
Dr. Azeddine Beghdadi
Professor, University Sorbonne Paris Nord
Abstract:
The major upheaval in the scientific landscape over the last two decades due to the adoption of the deep learning paradigm as a universal approach to solving a multitude of scientific problems in various fields raises several questions. It is indeed surprising to see the scientific community adopting a methodology that stands out from traditional approaches based on deductive reasoning, observation/experiments, mathematical modelling and explainable problem solving. In this talk I will give my views on what has led to the gradual abandonment of traditional approaches in favor of the new trend of purely artificial intelligence-oriented scientific problem-solving. I will present my point of view through some key scientific questions and a few real-world applications in various fields that I have tackled during my career.
Through a few examples illustrating concrete cases in the field of computer vision, I will show, in the first part of my talk, the limitations of traditional approaches which could more or less explain the reasons which have slowed down their adoption these two decades. I will also give some thoughts on how to reconcile the two approaches (traditional approaches.vs.AI-based approaches) to make more progress in solving real problems and thus make DL-based approaches explainable.
In the second part of this talk I will present intelligent video surveillance as a multifaceted application and will discuss one of its critical aspects "perceptual quality" and its impact on the performance of some computer vision tasks. This talk will also be an opportunity to open up a debate on the subject and propose some avenues for reflection.
Biography:
Dr. Azeddine Beghdadi is Professor at University Sorbonne Paris Nord. He started his education at ENS Polytechnique and the Institute of Physics at University Es Senia (Oran, Algeria). He joined the graduate engineering school ENSTelecom in Paris in 1980. He received his Master in Optics and Signal Processing from University Paris-Saclay in June 1983 and the PhD in Optics and Signal Processing from Sorbonne University Pierre and Marie in June 1986. Dr. Beghdadi worked at different places, including « Laboratoire d’Optique des Solides » (CNRS, Sorbonne University) and LPMTM CNRS Laboratory (University Sorbonne Paris Nord). He is a founding member of the Laboratory of Information Processing and Transmission (L2TI laboratory) and was its director from 2010 to 2016. Dr. Beghdadi is the founder and Steering Committee Chair of the European Workshop on Visual Information Processing (EUVIP). He has co-authored more than 350 peer-reviewed journal and conference papers including 6 best paper awards and has given over 30 invited and plenary talks. He supervised more than 40 PhD/Post-Doc and around 35 master students. He has been involved, as LPI or PI, in more than 10 large-scale projects at national, European or international level. His main research interests include image/video quality enhancement and assessment, computational visual perception models for intelligent image/video processing and scene understanding and medical imaging. Dr Beghdadi is senior associate editor of IEEE Transactions on Image processing, associate editor of IEEE Transactions on Multimedia, “Signal processing: Image Communication” and Journal, Elsevier, European journal on image and video processing, Springer Verlag. Dr Beghdadi is a IEEE-MMSP technical management committee member (since 2021), elected member of the EURASIP Technical Area Committee – VIP (2021-2023).
Hope to see you all there.
All faculty, researchers and graduate students are invited to attend.