Thesis Defense of Abhishek Djeachandrane
Abhishek Djeachandrane, a PhD candidate from the CIR team, will defend his thesis on December 19, 2023, in the RT Amphitheater at the Vitry-sur-Seine campus of UPEC—120 rue Paul Armangot, 94400 Vitry-sur-Seine.
Title: Implementation of an Intelligent Decision-Support System Self-Regulated by Application-Centric Quality of Experience
Thesis Supervisor(s): MELLOUK Abdelhamid
Abstract:
In smart cities, video surveillance is an essential tool for ensuring public safety. In the past, security was enhanced by installing more cameras and centralizing their control. However, as the number of cameras increased, it became impossible for humans to manually monitor all footage in real time. Humans are prone to distraction and cannot maintain focus for extended periods. To address this challenge, experts have developed models capable of automatically detecting abnormal situations by analyzing video data and classifying it as normal or abnormal. These computer vision techniques have significantly advanced anomaly detection. However, due to the constantly changing and evolving nature of environments, conventional methods may not be sufficient to meet all the requirements of a real-world scenario. A literature review of "end-to-end urban video surveillance systems for asymmetric threats" was conducted to explore the subject in depth.
To address this issue, a development platform was meticulously designed, incorporating three fundamental strategies. First, it uses a corrective signal that considers exogenous, endogenous, and human factors in the surrounding context, known as "task-specific quality of experience." Second, it promises predictive systems based on machine learning and situational awareness to enhance system capabilities and performance outcomes. Modular approaches to customized learning schemes were explored, converging on a solution called "similarity-based reinforcement meta-learning" for multi-instance anomaly detection. Finally, the study recommends the adoption of self-managing systems that rely on autonomous computing principles for configuration, protection, and learning, based on machine learning, descriptive and inferential statistics, and control theory.
Together, these strategies provide a comprehensive and robust framework to address crucial questions using cutting-edge technologies and methodologies. By combining data enrichment, situational awareness, and autonomous computing, the final system effectively meets the needs of modern enterprises, for which the ability to learn, infer, and adapt quickly is as vital as the ability to be aware of the surrounding context.