Controle de Estilos Musicais em Tarefa de Aprendizagem – Design, Implementação e Avaliação Empírica
Resumo
Resumo: O presente estudo propõe, implementa e avalia a seleção inteligente de estilos musicais com Aprendizado por Reforço. A avaliação é empírica e realizada no contexto de experimentação humana com encadeamento, uma tarefa tipicamente estudada em Psicologia. Valores médios de desempenho coletivo (número de tentativas, duração da sessão e latência da resposta) em três condições experimentais (silêncio, aleatório e inteligente), na primeira e na quarta seqüências programadas em cada condição e entre condições, revelaram reduções significantes. Para evitar transições musicais bruscas, um anel unidimensional foi treinado baseado em Mapas Auto-organizáveis de Kohonen para representar adequadamente os estilos musicais.
Abstract: The present study proposes, implements, and evaluates the intelligent selection of musical styles with Reinforcement Learning. The evaluation is empirical and undertaken in the context of human experimentation with chaining, a typical psychological task. Average values of group performance (number of trials, session duration and response latency) in three experimental conditions (silence, random and intelligent), at the first and fourth programmed sequences in each condition and among conditions have revealed significant reductions. In order to avoid “crisp” musical transitions, a (one-dimensional) ring is trained based on Kohonen’s Self- Organizing Maps and on a proper representation of musical styles.
Abstract: The present study proposes, implements, and evaluates the intelligent selection of musical styles with Reinforcement Learning. The evaluation is empirical and undertaken in the context of human experimentation with chaining, a typical psychological task. Average values of group performance (number of trials, session duration and response latency) in three experimental conditions (silence, random and intelligent), at the first and fourth programmed sequences in each condition and among conditions have revealed significant reductions. In order to avoid “crisp” musical transitions, a (one-dimensional) ring is trained based on Kohonen’s Self- Organizing Maps and on a proper representation of musical styles.
Texto completo:
PDFDOI: https://doi.org/10.5753/cbie.sbie.2006.557-566