Name: Reuven Wachtel Seminar Description: Thermal facial recogniton has many upsides as a security measure - especially at night. Recent works have shown that using domain transfer, we can achieve better results, but is it really worth it Name: Roy Mahlab Title: EveryGraph: Task Agnostic Graph Neural Architecture

Name: Reuven Wachtel Degree: M.Sc. Supervisor: Yitzhak Yitzhaky Seminar Name: Thermal Facial Recognition Using Domain Transfer - Is it Worth it? Seminar Description: Thermal facial recogniton has many upsides as a security measure - especially at night. Recent works have shown that using domain transfer, we can achieve better results, but is it really worth it? Name: Roy Mahlab Degree: Electrical and computer engineering Supervisor: Dr Chaim Baskin Title: EveryGraph: Task Agnostic Graph Neural Architecture Abstract: Graph Neural Networks (GNNs) have demonstrated remarkable performance across a range of applications. However, their generalization across diverse tasks and datasets remains a fundamental challenge. Inspired by the success of foundation models in language, vision, and audio domains, recent efforts have sought to bring similar capabilities to graph-based learning. We introduce EveryGraph, a task-agnostic GNN architecture that generalizes across more than 20 benchmark datasets spanning link prediction, node classification, and graph classification. Remarkably, EveryGraph achieves this broad generalization by training on a single dataset for a single task—while requiring over $10 \times$ fewer computational resources than fully-supervised GNNs with comparable performance.