Federated learning in Tiny Machine-Learning
Research Groups VLSI THEORY, DESIGN AND APPLICATIONS (VLSILAB)
Thesis type THEORETICAL AND EXPERIMENTAL
Description The growing interest in Internet of Things and mobile Artificial Intelligence/Machine Learing applications is pushing the investigation on Deep Neural Networks (DNNs) that can be implemented in small and low-energy devices. This approach is known as Tiny Machine Learning (TinyML)
TinyML is one of the fastest‐growing areas of Deep Learning and is defined as a fast growing field of machine learning technologies and applications including hardware (dedicated integrated circuits), algorithms and software capable of performing on‐device sensing and data analytics at extremely low power. A lot of research regarding TinyML is focused on how to compress models and reduce the inference latency but only little has been done for online training, especially in a privacy-preserving manner and individuals are nowadays more and more concerned with who has access to their personal data and where these data are being shared.
The aim of this project is to understand how to cope with the challenges and limitations encountered in the process of training on-device with a Federated Learning (FL) framework.
Federated learning is a promising and recently introduced training technique that bases learning on decentralized edge devices or servers, each of them holding local data samples, without exchanging them. This approach stands in contrast to traditional centralized machine learning techniques where all the local datasets are uploaded to one server.
Possible developments of this works are:
* the investigation of approaches capable of allowing the intersection of TinyML and FL
* the development of Matlab/Python (or any other high-level programming language) of a framework mixing TintML/FL
* the actual implementation on microcontrollers of a FL-based system
Required skills Acquaintance with digital programmable devices such as microcontrollers and/or FPGAs is of interest. Capability of master theoretical subjects (related to signal processing algorithms in particular) as well as good programming skills (C, C++ and/or Python) is also a desired pre‐requisite.
Deadline 14/06/2023 PROPONI LA TUA CANDIDATURA