Walter Stechele is a Professor at the Technical University of Munich (TUM), holding a position in the Department of Integrated Systems within the TUM School of Computation, Information and Technology. His research focuses on hardware-software co-design for neural networks, embedded systems optimization, and edge computing applications. Key areas include FPGA acceleration of convolutional neural networks (CNNs), quantization-aware training, and adversarial robustness in multi-bit networks. He also investigates sensor fusion, automotive imaging systems, and medical image registration techniques. His work bridges theoretical advancements in machine learning with practical deployment challenges on resource-constrained hardware. Notable contributions include methodologies like MATAR (multi-quantization-aware training) and HW-flow-fusion (inter-layer scheduling for CNN accelerators). His research addresses critical issues such as numerical stability in 8-bit Winograd convolutions and optimizing imaging through automotive windshields. Leveraging FPGA platforms, he explores energy-efficient implementations of binarized neural networks (BNNs) for tasks like driveable area detection and gesture recognition on edge devices. His publications frequently emphasize real-world applications in autonomous driving, robotics, and medical imaging, demonstrating a commitment to translating algorithmic innovations into deployable systems.
