A team made up of researchers from Vicomtech and EHU / UPV wins the Best Student Paper Award at KES2021

The team of researchers made up of Unai Elordi, Alvaro Bertelsen, Luis Unzueta, Nerea Aranjuelo and Jon Goenetxea from Vicomtech and Ignacio Arganda-Carreras from UPV / EHU has been awarded the Best Student Paper Award at the KES2021 Conference held in Poland.

This conference includes a wide range of topics related to intelligent systems and their applications, as well as emerging intelligent technologies. At KES2021, research papers have been debated and defended based on, among others, Knowledge-Based Systems, Cognitive Systems, Neural Networks, Genetic Algorithms and Evolutionary Computing, Hybrid Intelligent Systems, Knowledge Discovery and Data Mining, Data Analysis and Pattern Recognition, Machine Learning and Computational neuroscience.

The research work "Optimal deployment of face recognition solutions in a heteregeneous IoT platform for secure elderly care applications", object of the award, has been developed within the framework of the European Project SHAPES of the Horizon 2020 Program and supports the process of preparing the doctoral thesis of some of the signing researchers.

The main objective of the research carried out aims to respond to the challenge posed by the inclusion of facial recognition for the authentication and surveillance of Internet of Things platforms for the elderly. In this group, interaction capacities are reduced due to causes related to aging, there are also a wide range of interaction devices and the need to manage biometric data in a secure way. In this type of process, it is very important to guarantee the protection and privacy of the data by implementing human-friendly authentication mechanisms. There is no doubt that facial recognition is an attractive feature for this type of user, but the level of computer knowledge can be an obstacle to its integration, at the same time as the different levels of cognitive abilities. These circumstances advise avoiding repetitive input of usernames and passwords.

The focus of the work carried out is based on lightweight deep neural networks for secure recognition and guide users during interaction. An automated procedure selects the appropriate inference engine, model configurations, and batch  size, based on device characteristics while biometric data is homomorphically encrypted to preserve privacy.