École Nationale Supérieure de
Techniques Avancées

Abstract

This report presents a research internship carried out from April 8th to July 26th. This internship took place in Politecnico di Milano alongside my Erasmus exchange. More precisely, the internship was done inside the Dipartimento di Elettronica, In- formazione e Bioingegneria (DEIB). The main objective of the internship was to implement a neural network architecture said to be very effective for bloodpressure estimation from photoplethysmography(PPG) signals. An implementation has al- ready been done by previous students. My role is to continue on this foundation and go further. The architecture comes from a scientific paper [5] that describes a new model of convolutional neural network parallelized with handcrafted features. The pre-existing implementation is using Matlab, Python and the folowing database [2]. The use of a new database [3] allowed us to better understand the performance obtained. Then the study of the model’s complexity and implementation through the comparison with existing online system [4] helped us better understand how to improve the model to achieve better results. Through modifications and new archi- tectures, we didn’t improve significantly the results of the system but were able to delve deeper into the understanding of our implementation and its results.