Marc Fraile

PhD Graduate, Uppsala University

Uppsala Social Robotics Lab; Vi3; Department of Information Technology

Email: marc.fraile.fabrega@gmail.com
GitHub: https://github.com/MarcFraile
At UU: https://www.katalog.uu.se/profile/?id=N20-1393
my face

About Me

I recently obtained a PhD from Uppsala University (Sweden), in which I studied the application of modern Computer Vision (CV) to Social Signal Processing (SSP) and Human-Robot Interaction (HRI). My thesis, "Computer Vision and Explainability in Human-Human and Human-Robot Interaction", investigated the applicability of end-to-end Deep Learning techniques to predict social behaviour during guided play.

Since social interaction datasets obtained through controlled experiments tend to be very small, the research community typically prefers the use of pre-trained feature extractors, combined with shallow ML models. Me and my collaborators showed that in some cirumstances the end-to-end approach can be superior, though it can also fail if the social signals are sufficiently subtle. To support this analysis, we collected the UpStory dataset, a novel social interaction dataset in an educational context.

I also investigated the use of Explainable AI (XAI) in Computer Vision, and proposed an approach to objectively compare different "attention map" techniques. I further studied the use of XAI as an aid to help users of a Social Robot understand how the robot is working.

My project was funded by Uppsala University's Centre for Interdisciplinary Mathematics (CIM), and by the Swedish Research Council's ELECTRA project (grant no. 2020-03167). I was a member of the Uppsala Social Robotics Lab (USRL), as well as Methods for Image Data Analysis (MIDA).

Teaching

  • Master's thesis supervision: Harichandar, A. "Multimodal Prediction of rapport level in dyadic child-child interaction." Aditya Harichandar joined the Uppsala Social Robotics Lab as a student intern in Autumn 2023. Under my supervision, he curated an audio feature dataset based on the data collected in my paper "UpStory: the Uppsala Storytelling dataset", and performed rapport prediction using both feature-based methods (using scikit-learn) and audio-based methods (using PyTorch). He also worked on replicating my published video-based results, and exploring modality fusion variants to perform audiovisual prediction. His work culminated in the defense of his masters thesis in Spring 2024, in which I acted as the supervisor. (2024)
  • Intelligent Interactive Systems (1MD032 / 1MD039): Master's level course with a focus on the hands-on design and implementation of conversational agents. Duties: design of learning materials, teaching lab sessions, grading homework, supervising and grading final projects. Due to ongoing evolution of the field, I was asked to re-design and teach part of the course in three consecutive editions. (2021 - 2023)

Publications

  • Fraile, M., Varni, G., Lindblad, J., Sladoje, N., & Castellano, G. "Are We Friends? End-to-End Prediction of Child Rapport in Guided Play." To be published in the ECCV 2024 proceedings. (Workshop)
  • Fraile, M., Calvo-Barajas, N., Apeiron, A.S., Varni, G., Lindblad, J., Sladoje, N., Castellano, G. "UpStory: the Uppsala Storytelling dataset." arXiv:2407.04352. (Under Review) [DOI] [Code]
  • Zhong, M., Fraile, M., Castellano, G., & Winkle, K. "A Case Study in Designing Trustworthy Interactions: Implications for Socially Assistive Robotics." Frontiers in Computer Science 5: 1152532. (Journal) [DOI]
  • Fraile, M., Fawcett, C., Lindblad, J., Sladoje, N., & Castellano, G. "End-to-End Learning and Analysis of Infant Engagement During Guided Play: Prediction and Explainability." Proceedings of the 2022 International Conference on Multimodal Interaction. (Conference) [DOI] [Code]
  • Fraile, M., Lindblad, J., Fawcett, C., Sladoje, N., & Castellano, G. "Automatic analysis of infant engagement during play: An end-to-end learning and Explainable AI pilot experiment." Companion Publication of the 2021 International Conference on Multimodal Interaction. (Workshop) [DOI]