Marc-André Cossette
Marc-André Cossette [CA/QC] is a transdisciplinary Canadian artist who employs sound, visual, and interaction design in the creation of interactive installations and audiovisual performances. Marc-André holds a bachelor’s degree in interactive media and a master’s degree in experimental media (UQAM). He is currently pursuing a Ph.D. program at Concordia University in which he is exploring the use of artificial Intelligence and Artificial Life algorithms in stage performances. In addition to his research-creation practice, Marc-André is a member of ALMA with Alexandre Saunier, and has collaborated with several artists as a sound, visual, interaction, and stage designer. He is also the co-creator and co-host of the REC podcast series on research-creation produced in collaboration with CHOQ.fm and the Hexagram Network. His work has been presented both in Canada and internationally, notably at ELEKTRA, MUTEK, Tangente Danse (Canada), Ars Electronica Festival (Austria), ISEA Barcelona, El Matadero (Spain), Zanzibar International Film Festival (Tanzania), and CMMAS (Mexico).
Creating with artificial neural networks: from micro-generation to real-time composition agents
In this talk, I trace how neural networks were implemented in my work as a solo artist, with the audiovisual duo ALMA, and in collaboration with dance/circus performers. In doing so, I highlight the shift I operated in my practice from micro-generation with neural nets to the development of real-time composition agents.
What I call micro-generation refers to elements in micro time units generated by the neural net. These elements can be represented on stage by musical notes, shapes, colors, sound, or image transformations. In this scenario, I conceive the role of the human performer as organizing these streams of micro-generation, thus giving meaning to the performance.
In the upcoming performance “ab initio,” I work in collaboration with Alexandre Saunier, Paloma Leyton, and Bastian Miranda to reverse the practice of micro-generation I have followed for the past six years. To materialize this reversal, I train deep neural networks using imitation and reinforcement learning to organize, on the performance time scale, smaller units (musical motifs, light patterns, movements) that have been created by the artists involved in this project. These models will become real-time composition agents, tasked with making decisions that will impact the unfolding of the performance.
The talk thus offers an account of the transformation of the creative process involved in shifting from micro-generation to real-time composition agents, based on experiences, published artworks, and studio experimentations.
