This workshop offers a comprehensive exploration into the proficient implementation of Deep Neural Network (DNN) models, seamlessly transitioning from PyTorch to Field-Programmable Gate Arrays (FPGAs). Participants will navigate both software and hardware optimization realms, honing in on Quantization-Aware Training while harnessing the power of High-Level Synthesis (HLS) through the AXI stream interface for DNN model implementation. With a focus on essential tools such as Brevitas and FINN, the course content delivers a sophisticated understanding of efficient coding techniques tailored for optimal hardware acceleration.

Learning Outcomes

The workshop will give a knowledge boost around following points:

1. Quantization-Aware Training using reduced precision DNN parameters
2. Using AXI stream-based data flow architecture to implement the DNN model on FPGA

Course Content

Following topics will be covered in this workshop:

1. Introduction to High-Level Synthesis (HLS)
2. Introduction to Brevitas
3. Quantization-aware training of DNN models
4. Hardware Efficient Convolution Algorithm
5. Accelerating DNN models using FINN

Prof. Dr. Faisal Shafait is working as Professor in the School of Electrical Engineering and Computer Science (SEECS) at the National University of Sciences and Technology (NUST), Islamabad, Pakistan. He is also an Adjunct Professor at The University of Western Australia, Perth, Australia. His research interests include machine learning and computer vision with a special emphasis on applications in document image analysis and recognition. He is the director of Deep Learning Lab (DLL) at National Center of Artificial Intelligence (NCAI), NUST, Islamabad, Pakistan. He has received the IAPR/ICDAR Young Investigator Award by the International Association of Pattern Recognition (IAPR) in 2019 and have recently been included in the list of the World’s Top 2% Scientists compiled by Stanford University.