Description du projet :
By exploiting our knowledge gained on previous projects SoLiDA (advanced machine learning), ALDAAL (improved training for
machine learning) and Bühler (applied machine learning for image recognition), we will develop and exploit the active learning
capabilities to increase the performance and reliability of the results of machine systems.
We will develop a generic methodology for active learning that will be validated on practical use cases. Two use cases will be
considered in our project:
1. Rice grains images taken on a tray and classified as good or bad grains.
Human experts will interact with the system to detect errors, their corrections will be sent to the machine learning system
and used to improve the overall result. We will validate this solution with Buehler, a swiss company very active in the
sorting machines for food industry.
2. Vineyard aerial images to detect different objects and the lines of vine.
Human experts will help to identify objects that have to be recognized on an aerial image, then they will correct step by step
the results obtained by the machine learning system. We will validate this solution with pix4D and aero41, startups that are
creating and exploiting very precise aerial images for agricultural purpose.
We will develop a prototype application based on a web app that will allow very simple interaction with the human experts and
let them evaluate, in a very short period, their impact on the learning performances.
Equipe de recherche au sein de la HES-SO:
Genoud Dominique, Mayoraz Calixte, Zuber Lucien, Reichenbach Julien, Arbellay Olivier, Verma Himanshu, Chianella Nicolas, Treboux Jérôme
Partenaires académiques: VS - Institut Informatique de gestion
Durée du projet:
25.06.2019 - 10.12.2020
Montant global du projet: 157'050 CHF