RESEARCH

SEASIDE (SEarch, Analyze, Synthesize and Interact with Data Ecosystems)

Different features of human-centered data, such as heterogeneity, multidimensionality, as well as dynamic characteristics require the development of models, methods and software tools to efficiently process them (corpus creation, pattern extraction, classification, generation), while taking into account the nature of the involved media: gesture, or text. The research activities of the SEASIDE team aim to develop advanced solution for content-based indexation and retrieval (similarity search), and  to propose machine learning methods for classification, clustering and interactive control, based on different levels of representation.

We focus more specifically on three main thematic axis:

  • Data mining : this research axis is transversal. It addresses data processing (indexing, filtering, and retrieval), focusing on similarity measures and kernel-based methods for classification and control. The methods are applied to various types of data, including temporal series (for example motion data), sequences, or textual data. 
  • Analysis / synthesis / recognition of gestures: this thematic axis is centered on the study of human gesture. We are interested in the expressive quality of gestures, and in the biological modeling of sensorimotor behavior through machine learning approaches, both for the analysis and generation of new behaviors. Furthermore we aim at providing intuitive and efficient access modes in large volume of motion capture data banks. Among application domains are sign language and musical gesture analysis and synthesis.
  • Semantic representations: finally, we rely on higher level representations (semantic, phonological) to access more efficiently data (extraction of resources from annotated corpora), and to generate contents from scenarios (new utterances in sign language, gesture production from scores, etc.). Exploiting the coupling between digital and symbolic abstraction levels is indeed one of our objectives.