Ph.D. student under the direction of H. PAPADOPOULOS

Thesis title: Statistical Relational Learning for Music Information Retrieval
Thesis abstract: This PhD project aims to develop techniques for interacting with large data bases of music audio signals. These signals are complex both from a physical point of view (multiple sound sources, noisy observations, etc.) and a semantic point of view: they contain interdependent music information (melody, chords, rhythm, etc.). These two aspects have been treated separately so far. The probabilistic models developed for processing music signals take into account the uncertainty of the audio, but have limited relational structure. The approaches based on logic can describe a complex relational structure but are limited to symbolic representations of music. This thesis proposes to explore the formalism of Markov Logic Networks that combines these two aspects to describe the hierarchical metric structure of an audio signal of music.