Quantized Neural Networks with Xilinx PYNQ

QPYNQ: Workshop on

Quantized Neural Networks with Xilinx PYNQ

QPYNQ logo

Curious about FPGAs and deep neural networks? Want to use neural networks in your embedded application without consuming lots of power? Quantized Neural Networks (QNNs) deliver excellent recognition accuracy without costly floating-point operations, and are blazing fast and highly energy-efficient when implemented on FPGAs.

Join the QPYNQ workshop to learn more about QNNs, and get hands-on experience with deploying them on the Xilinx PYNQ-Z1 platform. Prior exposure to FPGAs and QNNs is not required. We will cover the basics and use overlays (prebuilt hardware libraries) linked with Jupyter Notebook to accomplish learning goals.



The QPYNQ workshop included three hour-long slots with breaks in between. Each slot included a lecture and hands-on practical exercises on the PYNQ-Z1 board. You can download the lecture slides below.

  1. A Hands-on Introduction to FPGAs and Overlays by Ananya Muddukrishna: FPGAs, PYNQ, and programmable overlays
  2. Quantized Neural Networks by Michaela Blott and Nick Fraser: computational benefits on FPGAs, how to train QNNs
  3. BNN-PYNQ Overlays by Yaman Umuroglu: using the BNN-PYNQ overlay for inference on the MNIST, CIFAR-10 and GTSRB datasets, setting up weights for new network (Fashion-MNIST)

Information and Registration


Date and Time

Wednesday September 20, 2017

13:15 - 17:15


RISE SICS, Kistagången 16, Stockholm

Bring Your Own Laptop

Participants need to bring their own laptops. All other necessary equipment and tools will be provided at the workshop.



RISE logo XILINX logo

NTNU logo