275 research outputs found
Pierrick Sorin. Comedy in Video Art
No full textThe article reveals the genre-stylistic specificity of artworks of the French video artist Pierrick Sorin, created between the 1980s and the 2020s. It studies the methods and techniques characteristic of P. Sorin’s work, particular attention being paid to his use of comic techniques. The novelty of the research lies in the consideration of P. Sorin’s creative work as an integral artistic system in its development. The relevance of the present study is determined by the need to expand ideas about the principles of shaping and the features of drama in the genres of contemporary art.
The article shows that P. Sorin developed the techniques and methods of video art in an original and spectacular way. Out of the creative methods used by the artist, the author of the article focuses on the following ones: the attraction of expressive means of other arts (cinema, circus, music hall, variety theatre, etc.) and the combination of modern technical and technological solutions with the style of comedy films, buffoonery, and clowning, which is manifested in the plots, expressive means, the specifics of the characters. The classification of P. Sorin’s works proposed by the author includes video art, optical theatres (miniature installations), stage performances and scenographic works.
The author comes to the following conclusions: the plot structure of P. Sorin’s works, depending on the genre, includes miniature scenes-sketches, parody films that repeat scenario designs of the originals, and series of sketches combined in a holistic performance by means of recapitulation. In his scenographic works, P. Sorin creates a comic discourse, a play with the viewer based on technical attractions. The comic techniques characteristic of P. Sorin’s work are: comic absurdity, illogical or impossible combination of elements, the method of inconsistency, the method of unexpected, repetition, and the use of taboo topics
Human mental states classification using EEG by means of Genetic Programming
No full textThe advances in the development of Brain-Computer Interfaces(BCI) have been increasing in recent years, mostly because the level ofconvergence from multi-disciplinary techniques has evolved. The electroencephalography(EEG), a brain recording method studied in thisthesis, allows the construction of BCIs, however the signals are rathercomplex to process, which requires methodologies that efficiently extractpatterns from them. This thesis explores two directions: first, a systemis proposed for the epilepsy seizures recognition using a combinationof signal processing methods for an efficient feature extraction; second,it explores the usage of a meta-heuristic algorithm, namely GeneticProgramming (GP), as an alternative in the design of BCIs. Nonetheless,there is currently open-issues in GP that this thesis also explores: is therea more efficient search methodology in the exploration by GP?; what isa proper representation depending on the studied problem?; which arethe most adequate search operators?. For the first topic, a thoroughlystudy is presented by introducing a memetic GP applied to regressionproblems. Then, it is extended by adapting it to classification problems.The results are positive; GP is greatly benefited from the combinationof a general and a Local Search (LS) methodology. The last two topicsare studied simultaneously in the development of a recognition systemfor mental states using EEG. A GP version (+FEGP) is proposed thatevolves feature extraction models by using specialized search operators,individuals representation and fitness function. The results show thatthe combination of these reaches a state-of-the-art accuracy for the particulartask of mental states recognition.Los avances en el desarrollo de Interfaces Cerebro-Computadora (BCI, porsus siglas en inglés Brain-Computer Interface) se han incrementado en años recientes,principalmente porque ha evolucionado el nivel de convergencia detécnicas multidisciplinarias. La electroencefalografía (EEG), una teécnica degrabación de señales cerebrales estudiado en esta tesis, permite la construcciónde BCIs, sin embargo las señales son complejas para procesar, lo cual requieremetodologías que extraigan patrones de forma eficiente. Esta tesis explora dostópicos principales: primero, se propone un sistema para el reconocimientode convulsiones de epilepsia usando una combinacion de metodos de procesamientode señales para la extracción eficiente de rasgos; segundo, explora eluso de un algoritmo meta-heurístico, Programación Genética (GP, por sus siglasen inglés Genetic Programming), como una alternativa en el diseño de BCIs.Sin embargo, existen temas sin resolver in GP que esta tesis explora: ¿existe unametodologia de búsqueda en GP más eficiente?; ¿cual es una representaciónapropiada dependiendo del problema a estudiar?; ¿cual son los operadores debúsqueda más adecuados?. De esta forma, se presenta un estudio a fondo conla introducción de un GP memetico aplicado a problemas de regresión. Despues,se extiende adaptandolo a problemas de clasificación. Los resultados sonpositivos; GP se beneficia fuertemente de la combinación de una metodologiageneral de busqueda y una local (LS, por sus siglas en inglés Local Search). Losultimos dos cuestionamientos se estudian simultáneamente en el desarrollo deun sistema de reconocimiento para estados mentales usando EEG. Se proponeuna versión de GP (+FEGP) que evoluciona modelos de extracción de rasgosusando operadores especializados de busqueda, representación de individuosy función de aptitud. Los resultados muestran que esta combinación permiteuna exactitud de clasificación que aporta en el estado-del-arte para la tarea particulardel reconocimiento de estados mentales
Human mental states classification using EEG by means of Genetic Programming
No full textThe advances in the development of Brain-Computer Interfaces(BCI) have been increasing in recent years, mostly because the level ofconvergence from multi-disciplinary techniques has evolved. The electroencephalography(EEG), a brain recording method studied in thisthesis, allows the construction of BCIs, however the signals are rathercomplex to process, which requires methodologies that efficiently extractpatterns from them. This thesis explores two directions: first, a systemis proposed for the epilepsy seizures recognition using a combinationof signal processing methods for an efficient feature extraction; second,it explores the usage of a meta-heuristic algorithm, namely GeneticProgramming (GP), as an alternative in the design of BCIs. Nonetheless,there is currently open-issues in GP that this thesis also explores: is therea more efficient search methodology in the exploration by GP?; what isa proper representation depending on the studied problem?; which arethe most adequate search operators?. For the first topic, a thoroughlystudy is presented by introducing a memetic GP applied to regressionproblems. Then, it is extended by adapting it to classification problems.The results are positive; GP is greatly benefited from the combinationof a general and a Local Search (LS) methodology. The last two topicsare studied simultaneously in the development of a recognition systemfor mental states using EEG. A GP version (+FEGP) is proposed thatevolves feature extraction models by using specialized search operators,individuals representation and fitness function. The results show thatthe combination of these reaches a state-of-the-art accuracy for the particulartask of mental states recognition.Los avances en el desarrollo de Interfaces Cerebro-Computadora (BCI, porsus siglas en inglés Brain-Computer Interface) se han incrementado en años recientes,principalmente porque ha evolucionado el nivel de convergencia detécnicas multidisciplinarias. La electroencefalografía (EEG), una teécnica degrabación de señales cerebrales estudiado en esta tesis, permite la construcciónde BCIs, sin embargo las señales son complejas para procesar, lo cual requieremetodologías que extraigan patrones de forma eficiente. Esta tesis explora dostópicos principales: primero, se propone un sistema para el reconocimientode convulsiones de epilepsia usando una combinacion de metodos de procesamientode señales para la extracción eficiente de rasgos; segundo, explora eluso de un algoritmo meta-heurístico, Programación Genética (GP, por sus siglasen inglés Genetic Programming), como una alternativa en el diseño de BCIs.Sin embargo, existen temas sin resolver in GP que esta tesis explora: ¿existe unametodologia de búsqueda en GP más eficiente?; ¿cual es una representaciónapropiada dependiendo del problema a estudiar?; ¿cual son los operadores debúsqueda más adecuados?. De esta forma, se presenta un estudio a fondo conla introducción de un GP memetico aplicado a problemas de regresión. Despues,se extiende adaptandolo a problemas de clasificación. Los resultados sonpositivos; GP se beneficia fuertemente de la combinación de una metodologiageneral de busqueda y una local (LS, por sus siglas en inglés Local Search). Losultimos dos cuestionamientos se estudian simultáneamente en el desarrollo deun sistema de reconocimiento para estados mentales usando EEG. Se proponeuna versión de GP (+FEGP) que evoluciona modelos de extracción de rasgosusando operadores especializados de busqueda, representación de individuosy función de aptitud. Los resultados muestran que esta combinación permiteuna exactitud de clasificación que aporta en el estado-del-arte para la tarea particulardel reconocimiento de estados mentales
Prediction Performance and Problem Difficulty in Genetic Programming
No full textThe estimation of problem difficulty is an open issue in Genetic Programming(GP). The goal of this work is to generate models that predictthe expected performance of a GP-based classifier when it is applied toan unseen task. Classification problems are described using domainspecificfeatures, some of which are proposed in this work, and thesefeatures are given as input to the predictive models. These models arereferred to as predictors of expected performance (PEPs). We extendthis approach by using an ensemble of specialized predictors (SPEP),dividing classification problems into groups and choosing the correspondingSPEP. The proposed predictors are trained using 2D syntheticclassification problems with balanced datasets. The models are thenused to predict the performance of the GP classifier on unseen realworlddatasets that are multidimensional and imbalanced. This workis the first to provide a performance prediction of a GP system on testdata, while previous works focused on predicting training performance.Accurate predictive models are generated by posing a symbolic regressiontask and solving it with GP. These results are achieved by usinghighly descriptive features and including a dimensionality reductionstage that simplifies the learning and testing process. The proposed approachcould be extended to other classification algorithms and usedas the basis of an expert system for algorithm selection.La estimación de la dificultad de problemas es un tema abierto en Programación Genética (GP). El objetivo de este trabajo es generar modelosque puedan predecir el desempeño esperado de un clasificador basado en GP cuando este es aplicado a tareas de prueba. Los problemasde clasificación son descritos usando características de un dominio específico, algunas de las cuales son propuestas en nuestro trabajo y estascaracterísticas son dadas como entrada a los modelos predictivos. Nos referimos a estos modelos como predictores de desempeño esperado(PEPs, por sus siglas en inglés). Extendimos este enfoque usando un ensemble de predictores especializados (SPEPs, por sus siglas eninglés), dividiendo problemas de clasificación en grupos específicos y elegimos su correspondiente SPEP. Los predictores propuestos son entrenados usando problemas de clasificación sintéticos de 2D con conjunto de datos balanceados. Los modelos son entonces usados para predecir el desempeño de un clasificador de GP en problemas del mundo real antes no vistos los cuales son multidimensionales y desbalanceados. Ademas, este trabajo es el primero en proveer una predicción de rendimiento para un clasificador de GP sobre datos de prueba, mientras en trabajos previos se han enfocado en predecir el rendimiento para datos de entrenamiento. Por lo tanto, planteados como un problema de regresión simbólica son generados modelos predictivos exactos los cuales son resueltos con GP. Estos resultados son alcanzadosusando características altamente descriptivas e incluyendo un paso de reducción de dimensiones el cual simplifica el proceso de aprendizaje yprueba. El enfoque propuesto podría ser extendido a otros algoritmos de clasificación y usarlo como base de un sistema experto de selecciónde algoritmos
Prediction Performance and Problem Difficulty in Genetic Programming
No full textThe estimation of problem difficulty is an open issue in Genetic Programming(GP). The goal of this work is to generate models that predictthe expected performance of a GP-based classifier when it is applied toan unseen task. Classification problems are described using domainspecificfeatures, some of which are proposed in this work, and thesefeatures are given as input to the predictive models. These models arereferred to as predictors of expected performance (PEPs). We extendthis approach by using an ensemble of specialized predictors (SPEP),dividing classification problems into groups and choosing the correspondingSPEP. The proposed predictors are trained using 2D syntheticclassification problems with balanced datasets. The models are thenused to predict the performance of the GP classifier on unseen realworlddatasets that are multidimensional and imbalanced. This workis the first to provide a performance prediction of a GP system on testdata, while previous works focused on predicting training performance.Accurate predictive models are generated by posing a symbolic regressiontask and solving it with GP. These results are achieved by usinghighly descriptive features and including a dimensionality reductionstage that simplifies the learning and testing process. The proposed approachcould be extended to other classification algorithms and usedas the basis of an expert system for algorithm selection.La estimación de la dificultad de problemas es un tema abierto en Programación Genética (GP). El objetivo de este trabajo es generar modelosque puedan predecir el desempeño esperado de un clasificador basado en GP cuando este es aplicado a tareas de prueba. Los problemasde clasificación son descritos usando características de un dominio específico, algunas de las cuales son propuestas en nuestro trabajo y estascaracterísticas son dadas como entrada a los modelos predictivos. Nos referimos a estos modelos como predictores de desempeño esperado(PEPs, por sus siglas en inglés). Extendimos este enfoque usando un ensemble de predictores especializados (SPEPs, por sus siglas eninglés), dividiendo problemas de clasificación en grupos específicos y elegimos su correspondiente SPEP. Los predictores propuestos son entrenados usando problemas de clasificación sintéticos de 2D con conjunto de datos balanceados. Los modelos son entonces usados para predecir el desempeño de un clasificador de GP en problemas del mundo real antes no vistos los cuales son multidimensionales y desbalanceados. Ademas, este trabajo es el primero en proveer una predicción de rendimiento para un clasificador de GP sobre datos de prueba, mientras en trabajos previos se han enfocado en predecir el rendimiento para datos de entrenamiento. Por lo tanto, planteados como un problema de regresión simbólica son generados modelos predictivos exactos los cuales son resueltos con GP. Estos resultados son alcanzadosusando características altamente descriptivas e incluyendo un paso de reducción de dimensiones el cual simplifica el proceso de aprendizaje yprueba. El enfoque propuesto podría ser extendido a otros algoritmos de clasificación y usarlo como base de un sistema experto de selecciónde algoritmos
Genetic Programming Based on Novelty Search
No full textNovelty Search (NS) is a unique approach towards search and optimization,where an explicit objective function is replaced by a measureof solution novelty. However, NS has been mostly used in evolutionaryrobotics, its usefulness in classic machine learning problems has beenunexplored. This thesis presents a NS-based Genetic Programming(GP) algorithms for common machine learning problems, with the followingcontributions. It is shown that NS can solve real-world classification,clustering and symbolic regression tasks, validated on realworldbenchmarks and synthetic problems. These results are madepossible by using a domain-specific behavior descriptor, related to theconcept of semantics in GP. Moreover, two new versions of the NS algorithmare proposed, Probabilistic NS (PNS) and a variant of MinimalCriteria NS (MCNS). The former models the behavior of each solutionas a random vector and eliminates all the NS parameters while reducingthe computational overhead of the NS algorithm; the latter uses astandard objective function to constrain and bias the search towardshigh performance solutions. The thesis also discusses the effects of NSon GP search dynamics and code growth. Results show that NS can beused as a realistic alternative for machine learning, and particularly forGP-based classification
Real-time simulation of nonlinear audio effects using artificial intelligence
No full textCertains produits du domaine de la technologie musicale présentent des caractéristiques sonores uniques et recherchées par les musiciens. Ces caractéristiques sont souvent dues aux non-linéarités de leurs circuits électroniques. Nous cherchons à préserver le son de ces appareils par le biais de simulations numériques et à les rendre largement accessibles à de nombreux musiciens. Ce domaine d’étude a connu une forte augmentation de l’utilisation des réseaux de neurones pour la simulation au cours des dernières années. Dans ce travail, nous proposons d’utiliser les réseaux neuronaux pour cette tâche. En particulier, nous nous concentrons sur les méthodes de boîte noire capables de fonctionner en temps réel pour la modélisation des effets non linéaires, tout en gardant les contraintes du guitariste à l’esprit. Nous couvrons l’état actuel des connaissances et identifions les domaines qui méritent d’être améliorés ou étudiés, avec pour objectif final le développement d’un produit. La première étape, qui consiste à identifier les architectures capables de traiter en temps réel et en continu, est suivie par l’augmentation et l’amélioration de ces architectures et de leur pipeline d’apprentissage grâce à un certain nombre de méthodes. Ces méthodes comprennent l’intégration continue avec des tests unitaires, l’optimisation automatique des hyperparamètres et l’utilisation de l’apprentissage par transfert. Un prototype en temps réel utilisant un backend C++ personnalisé est créé à l’aide de ces méthodes. Une étude sur l’anticrénelage en temps réel pour les modèles boîte noire est présentée, car il a été constaté que ces réseaux présentent de grandes quantités de distorsion d’anticrénelage. Le travail sur l’incorporation du contrôle de l’utilisateur a également commencé pour une simulation complète des systèmes analogiques. Cela permet à l’utilisateur final de disposer d’une gamme complète de possibilités de modification de la tonalité. Les performances des approches présentées sont évaluées de manière objective et subjective. Enfin, un certain nombre d’orientations possibles pour les travaux futurs sont également présentées.Certain products in the realm of music technology have uniquely desirable sonic characteristics that are often sought after by musicians. These characteristics are often due to the nonlinearities of their electronic circuits. We are concerned with preserving the sound of this gear through digital simulations and making them widely available to numerous musicians. This field of study has seen a large rise in the use of neural networks for the simulation in recent years. This work applies neural networks for the task. Particularly, we focus on real-time capable black-box methods for nonlinear effects modelling, with the guitarist in mind. We cover the current state-of-the-art and identify areas warranting improvement or study with a final goal of product development. A first step of identifying architectures capable of real-time processing in a streaming manner is followed by augmenting and improving these architectures and their training pipeline through a number of methods. These methods include continuous integration with unit testing, automatic hyperparameter optimisation, and the use of transfer learning. A real-time prototype utilising a custom C++ backend is created using these methods. A study in real-time anti-aliasing for black-box models is presented as it was found that these networks exhibit high amounts of aliasing distortion. Work on user control incorporation is also started for a comprehensive simulation of the analogue systems. This enables a full range of tone-shaping possibilities for the end user. The performance of the approaches presented is assessed both through objective and subjective evaluation. Finally, a number of possible directions for future work are also presented
Genetic Programming Based on Novelty Search
No full textNovelty Search (NS) is a unique approach towards search and optimization,where an explicit objective function is replaced by a measureof solution novelty. However, NS has been mostly used in evolutionaryrobotics, its usefulness in classic machine learning problems has beenunexplored. This thesis presents a NS-based Genetic Programming(GP) algorithms for common machine learning problems, with the followingcontributions. It is shown that NS can solve real-world classification,clustering and symbolic regression tasks, validated on realworldbenchmarks and synthetic problems. These results are madepossible by using a domain-specific behavior descriptor, related to theconcept of semantics in GP. Moreover, two new versions of the NS algorithmare proposed, Probabilistic NS (PNS) and a variant of MinimalCriteria NS (MCNS). The former models the behavior of each solutionas a random vector and eliminates all the NS parameters while reducingthe computational overhead of the NS algorithm; the latter uses astandard objective function to constrain and bias the search towardshigh performance solutions. The thesis also discusses the effects of NSon GP search dynamics and code growth. Results show that NS can beused as a realistic alternative for machine learning, and particularly forGP-based classification
Le Katajjaq inuit au-delà de la patrimonialisation, Enquêter dans un terrain ouvert et vivant, être attentif aux fragilités
No full textInternational audienceIn this article, Pierrick Lefranc explores katajjaq, an Inuit vocal practice alternately perceived as "throat singing," a game, or a shamanic practice, and examines its modes of heritage-making and contemporary uses. The author traces the evolution of ethnomusicological studies dedicated to katajjaq, highlighting its political, epistemic, and aesthetic issues, as well as the tensions between heritage-making and (re)appropriation, in a context of epistemic justice and recognition of Inuit culture. The article highlights the limitations of traditional approaches, which tend to confine this practice within pre-existing categories such as music or playful activity. The author advocates for a living and collaborative investigative method, where katajjaq is studied through "specific usages" and its capacity to build bridges between cultural, social, and artistic worlds. Through concrete examples, such as its integration into contemporary music festivals, the article revisits the relationship between so-called traditional and contemporary practices while exploring the dynamics of cultural appropriation and reappropriation. The author emphasizes the importance of moving beyond logics of domination, identity recognition, and classification to uncover the transformative potential of this practice. In conclusion, the article proposes using katajjaq as a lever to rethink the interactions between research and creation, tradition and innovation, within a perspective of decolonizing knowledge. It invites us to celebrate the transformative power of cultural practices, both aesthetically and socially, and explores how research-creation can fully reveal their potential.Dans cet article, Pierrick Lefranc explore le katajjaq, une pratique vocale inuit perçue tour à tour comme un « chant de gorge », un jeu ou une pratique chamanique, et examine ses modes de patrimonialisation ainsi que ses usages contemporains. L'auteur retrace l'évolution des études ethnomusicologiques consacrées au katajjaq, en mettant en lumière ses enjeux politiques, épistémiques et esthétiques, ainsi que les tensions entre patrimonialisation et (ré)appropriation, dans un contexte de quête de justice épistémique et de reconnaissance de la culture inuit. L'article met en évidence les limites des approches usuelles, qui tendent à figer cette pratique dans des catégories préexistantes telles que la musique ou l’activité ludique. L'auteur appelle à une méthode d'enquête vivante et collaborative, où le katajjaq est étudié dans des "usages particuliers" et dans sa capacité à créer des ponts entre mondes culturels, sociaux et artistiques. À travers des cas concrets, comme son intégration dans des festivals de musique contemporaine, l'article revisite la relation entre pratiques dites traditionnelles et dites contemporaines, tout en explorant les dynamiques d’appropriation et de réappropriation culturelle. L'auteur souligne l’importance de dépasser les logiques de domination, de reconnaissance identitaire et de classification pour révéler les potentialités transformatives de cette pratique. En conclusion, l'article propose de faire du katajjaq un levier pour repenser les interactions entre recherche et création, tradition et innovation dans une perspective de décolonisation des savoirs. Il invite à célébrer le pouvoir de transformation des pratiques culturelles tant sur le plan esthétique que social et tentent de mieux comprendre comment la recherche-création peut permettre de mieux révéler leur potentialité
- …
