Machine learning for human behavioral discovery ; Apprentissage machine pour la découverte du comportement humain

The research is part of a multidisciplinary project that combines advances in computational science and in the humanities to understand and explain the role of visuomotor exploration strategiesin a climbing task during a learning protocol. We seek to model the dynamics of learning to under-stand how the frequency of novelty and the complexity of the learning situation affect the learningoutcome. Modeling from machine learning and human movement science has been used to designoptimal practice environments to train climbers to exploit adaptive behaviors that invite them to safely explore novel and functional patterns. This framework involves working with a behavioral signal that is a representation of the climber in the movement; this signal is multidimensional, has complex dynamics and has two main characteristics that limit its application in statistical learning: it is sparse (has missing measurements) and scarce in the number of samples. As a part of our work, in order to facilitate the creation of new qualitative metrics to assess the climbers’ performance, we first proposed a novel model for annotation of a behavioral signal trained on partially labelled sequences. This part of the thesis dealt with the first type of constraints. In the second part of the dissertation, we focused on adapting machine learning to evaluate the type of practice (control, variable and self-controlled) in order to apply a predictive modeling of transfer to compare them. In the pipeline design, we had to handle a small dataset (second type of constraints) to demonstrate higher predictive stability for self-controlled practice. ; Notre recherche s’inscrit dans le cadre d’un projet multidisciplinaire qui combine les avancées des sciences humaines et des sciences informatiques pour comprendre et expliquer le rôle des stratégies d’exploration visuo-motrice dans la tâche d’escalade lors d’un protocole d’apprentissage. Nous cherchons à modéliser la dynamique de l’apprentissage pour comprendre comment la fréquence de la nouveauté et la ...