Master Thesis in Emergent Width-Adaptive Representations via Spatially-Gated Channel Routing

Develop and implement a novel transformer architecture for autonomous driving perception that adaptively adjusts computational depth and width. Design and conduct experiments to assess model performance and efficiency for semantic segmentation and object detection, comparing results against baseline models. Analyze emergent network properties regarding resource allocation to image regions and internal feature representation organization. Contribute to cutting-edge research at the intersection of deep learning, computer vision, and autonomous driving to create safer, scalable perception systems. This three-to-six-month on-site thesis requires university enrollment and proficiency in Python and PyTorch.