Francesco Conti's main research interest is the deployment of advanced, brain-like deep learning based intelligence on top of ultra-low power, ultra-energy efficient programmable hardware. He is one of the main contributors of the open-source PULP platform project [http://pulp-platform.org] in particular for what concerns the integration of specialized hardware accelerators into the platform, but he is interested in the whole technological stack "from C to silicon".
His activity is currently focusing on the following themes, for which he is also available for tutoring students for major-effort theses:
- energy efficient hardware architecture for deep learning (e.g. digital design of specialized cores, accelerators, tape-out of research chips, FPGAs)
- software targeted at low power deep learning / inference on embedded devices (commercial, e.g. STM32 microcontrollers, and research-based, such as PULP)
- training of approximated neural networks (quantized, binarized) targeting minimal energy consumption
- applications of ultra-low power deep learning (autonomous drones/UAVs, mini-bots)
