Work­shop: Ef­fi­cient Neural Net­work In­fe­rence on FPGAs

 |  Paderborn Center for Parallel Computing (PC2)

On the 17th April 2024, the second workshop on Efficient Neural Network Inference on FPGAs took place at Paderborn Center for Parallel Computing (PC2). The workshop was organized and hosted by PC2 in cooperation with the Computer Engineering Group of Paderborn University, both partners within the eki project.

The eki project was started in 2023 and is funded by the German Federal Ministry for the Environment, Nature Conservation, Nuclear Safety and Consumer Protection under the funding line "AI Lighthouses for Environment, Climate, Nature, and Resources". The main goal of this eki project is to increase the energy efficiency of deep neural network (DNN) inference in AI systems through approximate techniques and mapping to high-end FPGA systems.

This workshop focused on two main objectives: understanding the concepts of DNN and FINN, which is an open-source framework optimized for efficient DNN inference on FPGA accelerators, and using FINN in practical hands-on sessions on the FPGAs on Noctua 2. To explain the advanced concepts of FINN, two fundamental sessions of FPGAs and neural networks were presented in the beginning of the workshop. Afterwords, the participants were able to train a neural network by themselves, as well as transform the trained network into an FPGA design using prepared JupyterHub hands-on notebooks. The attendees could then execute this design on the AMD/Xilinx Alveo U280 datacenter FPGA cards that are available as part of the Noctua 2 cluster and see the improved performance for the trained DNN inference.

The hybrid workshop attracted about 40 participants from all over Germany, e.g. Berlin, Bielefeld, Darmstadt, Hannover, Münster, and Wuppertal.