Explore the application of Conal Elliott's ConCat infrastructure for implementing Deep Learning in this 42-minute conference talk from ACM SIGPLAN's FHPNC'23. Discover how ConCat enables automatic differentiation for gradient computation and transforms code to run on GPUs using the Accelerate framework. Learn about the integration of these components in an industrial setting, comparing the advantages over traditional Python-based frameworks. Delve into the challenges of combining ConCat and Accelerate, and gain insights into high-performance GPU-based Deep Learning implemented in plain Haskell code. Join speaker Michael Sperber as he unravels the intricacies of this innovative approach to Deep Learning with categories.