Agathe Balayn

I am a postdoctoral researcher in Computer Science and Human-Computer Interaction at the Fairness, Accountability, Transparency, and Ethics in AI group (FATE) at Microsoft Research.

My research focuses on uncovering the harms that deploying machine learning (ML) systems into society can raise, understanding their causes, and envisioning remedies. I am particularly interested in characterizing ML supply chains, and in developing practical tools and policy solutions to support ML practitioners. For that, relying on in-depth qualitative inquiries in the production and deployment environments of ML systems, I analyze ML using a social, organizational, and political-economical lens that remains informed by the technical realities of production. While I’m interested in any kind of machine learning systems, my research primarily focuses on the ones using image data –machine learning based computer vision– (or sometimes tabular data) to perform classification or regression tasks.

In my research, I aim at making contributions of varying nature using a mixed methodology.

• Empirical studies. I study how practitioners develop, deploy, and evaluate machine learning systems using various tools. This allows to identify patterns and limitations in existing practices.
• Methods and tools. I develop algorithmic methods (e.g., relying on explainability methods) to facilitate the identification of the potential harms of developed systems before they occur. I also create user-interfaces to make existing methods more usable by practitioners, and to study their actual use in practice. Besides, I would like to develop other kinds of tools (e.g., conceptual frameworks, documentations, etc.) to further support machine learning practitioners in their daily practice with regard to harms.
• Literature surveys. I perform rigorous surveys of the scientific literature, to understand harms and existing approaches towards assessing or mitigating these harms, and to identify possible necessary re-orientations of the research based on insights from the empirical studies.
• Policies. With the empirical and literature insights, I aim at reflecting on the needs for additional AI policies.

Before my postdoc at Microsoft, I was a postdoctoral researcher at the Faculty of Electrical Engineering, Mathematics and Computer Science, at Delft University of Technology, as well as at the Technology, Policy, and Management faculty (TPM) of the same university, and a visiting researcher at ServiceNow and at Trento University (Italy).

Prior to that, I had the opportunity to work on other machine learning -related topics pertaining to the automatic detection of hate speech, the automatic translation of Japanese sign language to Japanese, and to the human-guided control of a robotic arm.

I hold an MSc degree in Computer Science from the TU Delft, the Netherlands, and an engineering diploma (with its MSc degree equivalence) in Systems and Control from the ENSTA ParisTech, Institut Polytechnique de Paris, France.

You can find my CV here.