Bring Machine Learning into the Abyss of FPGAs with Vitis AI скачать в хорошем качестве

Bring Machine Learning into the Abyss of FPGAs with Vitis AI

In recent years, alongside the explosive rise of Large Language Models (LLMs) in the AI landscape, the embedded industry has followed its own trend going toward bringing MLcloser to the edge and introducing new challenges such as balancing performance and energyefficiency. In this context, FPGAs offer an interesting solution combining parallelism, reconfigurability and relatively low power consumption. Yet, how can we bring such complex, high-level ML algorithms into the “abyssal” world of hardware design languages (HDL)? And what are the limitations of this integration? In this talk, we will attempt to answer these questions through a hands-on exploration of AMD’s Vitis AI framework. We will walk through its workflow to highlight the advantages and limitations and briefly examine alternative solutions to this approach.