Abstract:
In this project, we aimed to enhance the computational efficiency and deployment feasibility of neural networks through mixed precision quantization. We implemented two quantization-aware training (QAT) methods. Our results demonstrated significant reductions in model bit-width assignments while maintaining accuracy comparable to full-precision models.
Speaker: Omar Lahyani
Short bio:
Omar Lahyani is a fifth-year engineering student at Ecole Polytechnique de Tunisie. During 2024, he worked as a research intern at Barcelona Supercomputing Center (BSC) to develop his final thesis and obtain his diploma with a project focused on efficient AI acceleration.
Speakers
Speaker: Omar Lahyani.
Host: Francesc Moll.