On January 30, at the Faculty of Electronic Engineering, within the AIDA4Edge project, we hosted a full-day program of lectures, hands-on training, and collaborative discussions.

The morning session began with a lecture on Neural Architecture Search (NAS) delivered by Riccardo Zese from the University of Ferrara, who introduced the main principles, methods, and research challenges of automated neural network design. Our project partner presented, in a very engaging way, the three key dimensions of NAS design that enable more efficient exploration of architectures and the identification of the right network shape that delivers superior DNN performance for multi-objective optimization problems.

As part of this lecture and during a subsequent hands-on tutorial, Riccardo Zese and Alice Bizzari presented Symbolic DNN-Tuner, a NAS and hyperparameter tuning framework developed by our partners at the University of Ferrara. Our team from the Faculty of Electronic Engineering contributed to the implementation of several post-training quantization methods, supporting the development of energy-efficient, high-accuracy AI systems suitable for deployment on edge devices. This tuner will be used for designing ANN models within the hybrid SNN–ANN architecture being developed in the project.

The afternoon session concluded with a research meeting, featuring productive discussions on ongoing research activities, the exchange of insights, and alignment on future project directions, particularly related to the design of the hybrid SNN–ANN pipeline.