Séminaire ARMEDIA présenté par Elena Vildjiounaite, PhD, lundi 17 avril 2023 14h, en H218 Evry

The SAMOVAR-ARMEDIA team is pleased to invite you to its research seminar on 

Challenges of behaviour-based stress detection in real life

presented by Elena Vildjiounaite, PhD, VTT Technical Research Centre of Finland

on Monday April 17th, at 2:00 PM

The seminar will take place in H218 (Etoile building) on Evry campus and will be broadcasted online at the following link :


 (we advise to use Google Chrome instead of Firefox to connect to the link)

Presenter: Elena Vildjiounaite, PhD, works in VTT Technical Research Centre of Finland. Her current tasks include project preparation and coordination, and leadership of work packages. Currently, Elena Vildjiounaite is international coordinator of Mad@Work project. During over 20 years at VTT Elena Vildjiounaite worked in multiple international and national R&D projects in the domains of biometrics, privacy protection, human-computer interaction, mental wellbeing, and monitoring of elderly at home.

Abstract: It is estimated that around 50% of all lost working days have some links with work stress. In addition, stress can cause presenteeism, employer turnover and early retirement, and replacement of highly qualified knowledge workers is costly. Typically, attitudes of the employees are measured by questionnaires once a year or two years, while about 31% of new hires leave their jobs within the first six months.

Mad@Work international (ITEA) project aims at developing methods to detect and mitigate stress of knowledge workers, suitable for long-term real life use on regular basis. This goal presents a lot of research challenges: first, systems should be convenient, privacy-preserving and accepted by the employees and the employers. Second, the majority of stress detection studies to date were performed in the labs, where stress is induced by for example asking test subjects to do mental arithmetic quickly. It is not exactly what we do at work; instead, methods to detect long-lasting troubles are needed. Third, the majority of stress detection studies to date used fully supervised stress detectors. Since stress is highly person-dependent, it means the requirement that each end user must provide 60-100 self-reports of the kind “today I was stressed” to train classifiers, that is, 3-4 months long data collection is required before the system can even start working!

This seminar will present Mad@Work approach and research challenges in more details, and will open a discussion for potential topics for cooperation.