Edge AI for Low-power Machine intelligence
The industrial chair is concerned with the problems related to the integration of AI on embedded targets. This goal is addressed by compressing deep convolutional neural network models during the training phase, the aim being to achieve embedded inference with an energy budget of less than a milliwatt and less than a MegaByte of memory.
The hardware targets are SoC solutions based on both microcontrollers and neural network engines developed by Dolphin Design and LEAT Lab respectively.
The ambition of the project is to address the issue of efficiency from the learning phase to the deployment on optimized hardware targets. Compression, quantization and distillation methods are therefore addressed with a hardware-aware orientation to take account of the target’s specificities.
We will show how this new 3IA Côte d’Azur Industrial Chair tool can help industrial and academic partners to complement each other in tackling ambitious scientific and technical challenges. Over several years, this type of project will lead to scientific publications, patents and software and hardware solutions that can be exploited to give industrial partners an economic advantage.
Dolphin Design is a French semiconductor company. Dolphin designs and markets not only key functions for integrated circuits (IP), but also complete
integrated circuits such as ASICs and SoCs. The company boasts unique know-how in the field of optimizing the energy efficiency of electronic circuits and develops new solutions for Edge AI applications.
LEAT Lab is the research lab on Electronic, Antennas and Telecommunication of University Côte d’Azur and is located at Sophia-Antipolis. The eBRAIN group from LEAT is specialized in Edge AI and is leaded by Benoît Miramond, chaire 3IA on bio-inspired AI. The team develops new technologies based on Spiking Neural Networks.
https://www.worldaicannes.com/fr/demo-sessions/659440cf691e71008190571c