Alessandro Cacciatore

KEYWORDS: preterm infants; convolutional neural networks; green AI; model compression 

Abstract 

Preterm infants, who every year represent more than one in ten newborns all over the world, are more likely to develop mild-to-severe neuro-motor impairments. Assessing the quality of their spontaneous motility is a very effective way to early diagnose such future outcomes. In order to assist clinicians in this assessment, research in the Deep Learning domain has utilized Convolutional Neural Networks (CNNs) which, however, are often more expensive (in terms of computational and hardware requirements ) than healthcare can afford.  To develop fairer technologies, we study how Model Compression, specifically Knowledge Distillation, can be used to train inexpensive models that can guarantee results that comply with the high accuracy required in the clinical practice of preterm infants’ movements monitoring.

• EMAIL: a.cacciatore1@unimc.it
• LINKEDIN PROFILE
• ORCID 0000-0003-4189-2749