Aktualny program
Aktualny program
Piątki, godz. 11:00, sala 44 (Laboratorium Mi^2)
- 12 czerwca 2026, Paulina Tomaszewska (Wydz. MiNI, PW), Role of spatial and temporal context in classification using deep neural networks
Context plays a fundamental role in human perception and decision-making, enabling holistic reasoning from multiple interacting cues. But do Deep Learning models behave in a similar manner? In this talk, I will present my work on inspecting Deep Learning classifiers through the lens of temporal and spatial context, with a particular emphasis on the latter. I will show that many popular explainable AI methods for vision models analyze image regions in isolation, overlooking relationships between them. To address this limitation, I introduce methods and benchmarks for assessing spatial reasoning at local, semi-global, and global scales. Digital pathology will serve as a main case study, demonstrating the importance of spatial context in Deep Learning-based medical image analysis.