Résumé
This study focuses on human synchronization in dance: how individuals adapt their movements when dancing to music with a partner. We analyze a dataset of dyadic dance, where pairs of individuals performed in four experimental conditions either listening to the same or different music and either seeing each other or not. The individuals' whole-body movements were recorded using 22 motion capture sensors. A key question we investigate is whether motion alone can reveal the condition in which the dancers were recorded. To address this classification problem, we use Spiking Neural Networks (SNNs). Due to their intrinsic ability to encode temporal dynamics, these models seem well suited for analyzing such motion time series data. Since SNNs process discrete spike events, an essential preprocessing step involves encoding continuous movement data into spikes. Classification accuracy is evaluated through various preprocessing steps, network architectures and hyperparameters. The results open perspectives for a biologically plausible modeling of human perception of synchronization.