When used with deep learning, the symbolic music modality is often coupled with language model architectures.To do so, the music needs to be tokenized, i.e. converted into a sequence of discrete tokens.This can be achieved by different approaches, as music can be composed of simultaneous tracks, of simultaneous notes with several attributes.Until now, the proposed tokenizations rely on small vocabularies of tokens describing the note attributes and time events, resulting in fairly long token sequences, and a sub-optimal use of the embedding space of language models.Recent research has put efforts on reducing the overall sequence length by merging embeddings or combining tokens.In this paper, we show that Byte Pair Encoding, a compression technique widely used for natural language, significantly decreases the sequence length while increasing the vocabulary size.By doing so, we leverage the embedding capabilities of such models with more expressive tokens, resulting in both better results and faster inference in generation and classification tasks.The source code is shared on Github 1 , along with a companion website 2 .Finally, BPE is directly implemented in MidiTok 3 , allowing the reader to easily benefit from this method.

Deep learning models are now core components of modern audio synthesis, and their use has increased significantly in recent years, leading to highly accurate systems for multiple tasks. However, this quest for quality comes at a tremendous computational cost, which incurs vast energy consumption and greenhouse gas emissions. At the heart of this problem are the standardized evaluation metrics used by the scientific community to compare various contributions. In this paper, we suggest relying on a multi-objective metric based on Pareto optimality, which considers equally the accuracy and energy consumption of a model. By applying our measure to the current state-of-the-art in generative audio models, we show that it can drastically change the significance of the results. We hope to raise awareness on the need to more systematically investigate the energy efficiency of high-quality models, in order to place computational costs at the center of deep learning research priorities.

This paper describes the implementation of an argumentation system used for participatory management of environmental protected areas, more precisely to model the decision of a park manager artificial agent.This implementation is based on a BDI agent architecture, namely the Jason/AgentSpeak framework/language.After introducing the principles of BDI architecture and of argumentation systems, we will detail how we model arguments within the BDI (Belief-Desire-Intention) architecture.Then, we present the argumentation-based model of deliberation and decision by the park manager agent as a case study.We show how our argument-based approach allows to model various cognitive profiles of park managers (more conservationist or more sensitive to social issues), through different knowledge bases.We show examples of decisions produced by the park manager agent and examples of traces of arguments used during deliberation, which could be a base for explaining decisions.Before concluding, we point out future directions, such as using argumentation as a basis for negotiation between various agents.

International audience

A challenge for pervasive computing is the seamless integration of computer support with users' activities in a very dynamic setting, with deep human and resource mobility. Portable devices such as notebooks, PDAs and smartphones are becoming more and more popular, as their computational power increases and their prices fall. Moreover, the quick spread of wireless networks has permitted the exchange of data in places as diverse as university campus, airports, coffee houses and family homes. In this paper we introduce campus, a framework that supports the development of multi-agent, context aware, pervasive computing applications. Campus is designed to provide the necessary infrastructure for ambience intelligence applications.

Ubiquitous computing systems can be regarded as open systems where heterogeneous and previously unknown entities may spontaneously interact, due to intrinsic mobility of users and their devices. In this highly dynamic and heterogeneous scenario, applications must be capable of accessing the appropriate instances of the required services in each visited network or region. Some services, however, must be made available only to users and applications that fulfill some conditions. In ubiquitous systems the interaction between client and server applications happens in a physical space, involving human and artificial agents that act under social and administrative rules. We propose the integration of context-awareness with a social regulation approach to control the interaction of users and their applications in such environments. This has the advantage that access policies and rights can be defined and monitored independently of the applications for the ubiquitous system.

International audience

This paper discusses the introduction of explicit metaclasses á la ObjVlisp into the Smalltalk-80 language. The rigidity of Smalltalk metaclass architecture motivated this work. We decided to implement the ObjVlisp model into the standard Smalltalk-80 system. The resulting combination defines the Classtalk platform. This platform provides a full-size environment to experiment with class-oriented programming by combining implicit metaclasses á la Smalltalk and explicit metaclasses á la ObjVlisp. Obviously, these experiments are not limited to the Smalltalk world and will be useful to understand and practice the metaclass concept advocated by modern object-oriented languages such as ObjVlisp and CLOS.

An object-oriented computation model is presented which is designed for modelling and describing a wide variety of concurrent systems. In this model, three types of message passing are incorporated. An overview of a programming language called ABCL/1, whose semantics faithfully reflects this computation model, is also presented. Using ABCL/1, a simple scheme of distributed problem solving is illustrated. Furthermore, we discuss the reply destination mechanism and its applications. A distributed “same fringe” algorithm is presented as an illustration of both the reply destination mechanism and the future type message passing which is one of the three message passing types in our computation model.