**Si vous souhaitez recevoir les mails liés aux Semidocs, vous pouvez vous inscrire sur cette liste de diffusion.**

# ** Présentations à venir **

##### Mercredi 6 avril 2021 à 15h - Amphithéâtre - LaBRI Bât A30

#### Vincent Martin - Image et Son - LaBRI - Université de Bordeaux

#### ** Titre : Pathology detection in voice: a pictorial journey of bad practices**

** Résumé : ** During a systematic review of the corpora used to detect different pathologies in the voice, I put aside a set of
figures, graphs, (almost) explanatory diagrams whose comprehension, even with the context, is often difficult.
This semidoc thus proposes a pictorial ballad among the worst figures encountered in more than 300 articles published in
journals and conferences of varying prestige, from the reference journal hosted by Elsevier to the worst predatory
journals.

# ** Historique des présentations **

##### Mercredi 6 avril 2021 à 15h - Amphithéâtre - LaBRI Bât A30

#### Philippe SWARTVAGHER - TADaaM - INRIA - Université de Bordeaux

#### ** Titre : About Interactions in Distributed Task-Based Runtime Systems for HPC Systems**

** Résumé : ** Super-computers dedicated to High Performance Computing applications are
becoming more and more powerful, yet more and more complex. This complexity
makes it difficult to use efficiently all the computing capacities these
machines can offer. To ease the development of applications targeting such
computers, task-based runtime systems are developed. They propose a programming
model where the programmers can focus on the code of their applications and
take benefit from the machine abstraction offered by the runtime system. On
super-computers, network communications can be a bottleneck for the general
application performances, hence they have to be carefully optimized. In my
thesis, I try to improve the interactions between task-based runtime systems
and communication libraries, to improve their collaboration.
During this presentation, I will give a quick introduction to HPC, present the
concept of task-based runtime systems and explain the main contributions of my
thesis.

##### Mercredi 23 mars 2021 à 15h - Amphithéâtre - LaBRI Bât A30

#### Séance Spéciale: Tell us how you do it - LaBRI - Université de Bordeaux

#### ** Tell us how you do it**

** Résumé : ** We know how wonderful all your presentations are. Now we want to know how you do it!
We organize a session in which
several people will share in 5 minutes some tips about how to bring to the next level the quality of your
slides,
posters or papers. If you know some nice latex packages, useful software, or simply want to share some advice,
please,
contact us!

##### Mercredi 16 mars 2021 à 15h - Amphithéâtre - LaBRI Bât A30

#### Maxime Poret - Image et Son - LaBRI - Université de Bordeaux

#### ** Titre : Towards a generic model for data sonification**

** Résumé : ** Sonification is a pluridisciplinary domain at the crossroads of Human-Computer
Interaction, psychoacoustics and sound
computing, which consists in procedurally translating data into non-verbal sounds for the purposes of monitoring
or
exploration. Despite extensive research into universal frameworks and models in the past few decades, ad-hoc
data-to-sound pipelines still tend to be developed from scratch each time a new sonification project is started.
The aim
of this thesis is to work towards a simplified, modular design model for the representation of all mappings
between data
dimensions and sound synthesis parameters. Based on a few case studies submitted by researchers from various
domains
(biology, environment, manufacturing), we study how our mapping choices fit into the model we propose, before
experimenting on its implementation in the ossia score musical sequencer.

##### Mercredi 24 novembre 2021 à 15h - Amphithéâtre - LaBRI Bât A30

#### Jonathon Narboni - CombAlgo - LaBRI - Université de Bordeaux

#### ** Titre : Kempe chains: coloring and recoloring **

** Résumé : ** Graph coloring problems are usually used to model resource allocation problems. They are one of the most studied types of problem in graph theory, and the four-color theorem is maybe the most famous result in the field. In 1852 Guthrie asked if a geographical map can always be colored with only 4 colors so that no adjacent countries receive the same color, and in 1879 Alfred Kempe claimed that he has a proof of this theorem. His proof was proven false eleven years after by Heawood, but it turned out that the main tool of his proof, Kempe chains, are one of the most fruitful concepts in graph coloring. Kempe chains allow one to transform a coloring into another coloring, and the first real proof of the four-color theorem (like many other results in graph coloring) heavily relies on Kempe chains. In this talk we discuss how Kempe chains can be used to bound the number of colors, moreover, we also discuss how Kempe chains can be used in a question of its own interest: coloring reconfiguration.

##### Mercredi 10 novembre 2021 à 15h - Amphithéâtre - LaBRI Bât A30

#### Joséphine Calandra - Image et Son - LaBRI - Université de Bordeaux

#### ** Titre : The cognitive algorithm: a music listening model and tool for music analysis **

** Résumé : ** The recent evolutions of systematic musicology taking into account cognitive sciences radically modify our apprehension of time in music. Therefore, the emergence of cognitive musicology and specifically a cognitive analysis has resulted in musicology. It has made it possible to offer a significant evolution in the modes of representation of musical forms and structures. The computer implementation of these new ideas would make it possible to renew the way of understanding digital data flows. We present the automatic computation of formal diagrams, which are representations of the listened music's paradigmatic analysis. We discuss then a hierarchical representation of formal diagrams computed simultaneously, where upper-level représentations -thus representing longer musical objects- are dynamically computed depending on lower-level diagrams.

##### Mercredi 27 octobre 2021 à 15h - Amphithéâtre - LaBRI Bât A30

#### Théo Matricon - M2F - LaBRI - Université de Bordeaux

#### ** Titre : Statistical Comparison of Algorithm PerformanceThrough Instance Selection **

** Résumé : ** Empirical performance evaluations, in competitions and scientific publications, play a major role in improving the state of the art in solving many automated reasoning problems, including boolean satisfiability (SAT), constraint programming and Bayesian network structure learning (BNSL). To empirically demonstrate the merit of a new solver usually requires extensive experiments, with computational costs of CPU years. This not only makes it difficult for researchers with limited access to computational resources to test their ideas and publish their work, but also consumes large amounts of energy. We propose an approach for comparing the performance of two algorithms: by performing runs on carefully chosen instances, we obtain a probabilistic statement on which algorithm performs best, trading off between the computational cost of running algorithms and the confidence in the result. We describe a set of methods for this purpose and evaluate their efficacy on diverse datasets from SAT, CSP and BNSL. On all these datasets, most of our approaches were able to choose the correct algorithm with about 95% accuracy, while using less than a third of the CPU time required for a full comparison; the best methods reach this level of accuracy within less than 15% of the CPU time for a full comparison.

##### Mercredi 13 octobre 2021 à 15h - Salle 178 - LaBRI Bât A30

#### Vincent Martin - Image et Son - LaBRI - Université de Bordeaux

#### ** Titre : How (and how not) to design a corpus for machine learning in psychiatry ? **

** Résumé : ** While the global effort in machine learning is focused on building ever more complex learning models, little attention is given to the data used to train these algorithms, and to their impact on the generalization that can be made from these systems. The goal of this talk is to develop, in interaction with the audience, a rigorous methodology for building databases using the example of voice drowsiness detection, and to show that with these, it is possible to design epistemologically robust and meaningful systems.