Orange SA

..Within the Orange Innovation Division, whose ambition is to take Orange innovation further and strengthen its technological leadership, the IT and Services / BIZZ Department brings together service innovation activities for companies in the fields data and AI, IoT, connectivity, collaboration, workspace, customer relations, etc. A third of the 300 employees are dedicated to Research activities. Within the IT and Services business direction you will join the Data Intelligence 4 Business team whose activities focus on the development of services based on artificial intelligence.

Your role is to carry out a post-doc on : "Ethics of AI in federated learning     Europe defends responsible AI and published in mid-April 2021 the "AI act" [2] aimed at legislating on high-risk AI. Societal concerns about discrimination, privacy, transparency, explainability and the responsibility of data scientists and companies are growing [3]. At the same time, technology continues to improve the performance and accuracy of models. More particularly, federated learning [4] allows to respond to concerns about privacy/confidentiality between different actors. However, at the same time, it raises complex and new ethical problems especially in the case of federated learning involving several companies willing to train a common model (cross-silo federated learning) among which fair participation of all [5]. Finally in France from 2022, mobile operators and internet service providers will have to indicate the carbon impact of their customers' digital activities and artificial intelligence is a significant element to take into account.   The objective of the post-doc is to propose and develop tools allowing the development of cross-silo federated AI models that are above all fair and have a low carbon footprint. You will tackle the problem of fairness due to training data [4] [2] [3] and the mechanism of training a model [4] [5] [6]. You will also focus on making models explainable and less energy consuming [11][12].  By taking as a framework the generic use case of a high-risk AI and taking into account the specificities related to cross-silo federated learning , you will propose tools (KPI and alert system) to report any potential bias and minimize the carbon footprint of the system as a whole (indicators and warning system on all components). planning 

Personal qualities desired by the position.  You have very good knowledge of AI model development and particularly in federated learning / deep learning, edge intelligence. You are ethically sensitive. You are autonomous and take initiatives. You appreciate teamwork.   Skills (scientific and technical) qualities desired by the position.  You have a PhD In artificial intelligence or data science obtained in the last two years. Professional experience internships, etc.)  Work carried out in the field (publications, report, participation in workshops). requested training.