PhD project in Image Reconstruction using FPGA-based Generative AI

The Centre for Electronics Frontiers (CEF), led by Regius Chair of Engineering Prof Prodromakis, brings together diverse and interdisciplinary expertise for transforming modern society through technology. Our ambition is to push the frontiers of electronics through emerging technologies, disrupting current ways of thinking by innovating advanced nano/biosensors, safe and efficient energy storage solutions and novel hardware for AI. We are offering prospective PhD students the opportunity to join our team, interested in devoting their passion for addressing some of the challenges we have identified. This project will also be supervised by Dr Cristian Sestito.

The project aims at building accelerators based on Field Programmable Gate Arrays (FPGAs) and suitable to deliver computer vision tasks through Generative AI. Generative Adversarial Networks (GANs) based on Convolutional Neural Networks (CNNs) are promising candidates in this direction: they exploit adversarial learning and feature extraction to execute a multitude of applications, including image dataset generation, image-to-image translation, face frontalisation. Specifically, the project targets deploying applications like this on FPGA-based Systems-on-Chip (SoCs) to be showcased in real-time systems, with an in-depth investigation on optimisation techniques to reach high throughput and low energy footprint (e.g., data quantisation and pruning). This will require preliminary training using software frameworks, like PyTorch or TensorFlow.

The required skills are as follows:

• Knowledge and expertise on FPGA design for AI using Verilog/VHDL (mandatory).
• Knowledge and expertise on training and testing CNNs using SW frameworks, like PyTorch or TensorFlow (mandatory).
• Basic knowledge on Systems-on-Chip based on FPGAs (desirable).
• Previous experience on using 3^rd party IP cores for vision applications (desirable).
• Previous experience on training generative AI models (desirable).