Post-doc: Markov constraints applied to music/text generation

Type : Post-doc

Employer : UPMC - Université Pierre et Marie Curie - Paris 6

Workplace : Sony CSL Paris - 6, rue Amyot, Paris 75005

Sony CSL Paris was founded in 1996 and is a small but booming research cell, focusing on four areas: Music Access and Creation, Adaptive Language Systems, Sustainable Development and Comic Computing.

Skill area : Computer science, graph theory

Years of experience : At least 2 years

Salary range : 44,000 € annual gross, negotiable

Contract type : 4-year contract

Posting date : 12/02/2013

Context: The goal of the FlowMachines project is both to define and capture the stylistic essence of a given corpus (of musical works or texts) and to develop technologies to realize a style given arbitrary user constraints. These technologies are the key for developing a novel generation of authoring tools in which users navigate in large combinatorial spaces with high-level, meaningful controls. The project includes the development of such prototype applications, and the study of their impact on real-world creation tasks in music and literature. Although very ambitious and challenging, the team has already produced proof-of-concepts and core technologies and demonstrations (see publications and references).

Research topic : The project focuses on the notion of “style” for temporal sequences, such as melodies or literary texts. The goal is to develop style-centered content generation applications that boost the creativity of the users. The key idea is to combine statistical representations of style, such as Markov processes, with constraints that represent structural properties of the generated sequences. Our team has introduced a new theoretical framework, Markov Constraints, which consists in reformulating Markov processes as finite-domain constraint satisfaction problems. The subject of this position is to extend Markov constraints to capture long-range stylistic and structural aspects of sequences and to develop efficient algorithms to solve the resulting generation problems. The application context will be music and text generation.

References:

P. Roy, F. Pachet: Enforcing Meter in Finite-Length Markov Sequences, submitted to AAAI 2013

G. Barbieri, F. Pachet, P. Roy, M. Degli Esposti: Markov Constraints for Generating Lyrics with Style, ECAI 2012

F. Pachet, P. Roy: Markov constraints: steerable generation of Markov sequences, Constraints 16(2), 2011

F. Pachet, P. Roy, G. Barbieri: Finite-Length Markov Processes with Constraints, IJCAI 2011

Background/skills : Combinatorial optimization, in particular global constraints. Good knowledge (or strong willingness to quickly acquire competences) in graph theory, global constraints, and operations research. Ability to work co-operatively and effectively in an international, multi- cultural environment.

Please, send application to: pachetcsl@gmail.com