This year Istanbul Technical University was the host site for EvoStar 2010, the main European conference on evolutionary computation. State of the art research in the application of nature inspired algorithms was presented in a number of fields including telecommunications, complex systems, environmental issues, finance, games, image and signal processing, machine intelligence, optimization, dynamics, transportation systems, and the focus here, art, music, and design. Each topic was organized as a conference within a conference although cross attendance was encouraged.
The EvoMUSART section, as it is called, included 16 papers selected in a competitive peer review process. Like the other papers from the conference these papers will appear in the journal Applications of Evolutionary Computation published by Springer.
While EvoMUSART is more technical than some other venues for generative art research aesthetic goals are nevertheless primary. For example Sah et al presented a system for creating animated photomosaics where, over time, a large number of very small images evolve an arrangement creating one large image. In the system presented by Dubbin and Stanley users can indirectly train neural networks by making interactive selections within an evolutionary system. This leads to the creation of animated dancers who can perform in realtime given music as an input.
This year a good number of music systems were presented. Shao et al presented Jive, a system that uses interactive evolution to compose generative functions, and then allows the use of a mouse or Wii-controller to interact with those functions for real-time performance. And a system that can take a bass line and compose harmonization in the form of 3 and 4 note chords was presented by De Prisco et al.
One of the earliest techniques applied to machine composition that is still viable today is the use of Markov processes. Put simply a Markov process selects the next note based on probabilities determined by the previous few notes. The problem with such systems is that they can sound reasonably good a few measures at a time, but lack coherence or aesthetic value over the full course of a piece. Davismoon and Eccles presented a system that overcomes this problem to some extent by adding musical constraints to the Markov probabilities.
In the realm of new synthesis techniques Sequera and Miranda presented a method where a cellular automata is mapped into a spectrogram that determines a timbre. Codognet and Pasquet reported on their Sound Agents system that creates an immersive ambient sound environment using 24 loudspeakers, a subwoofer, and multi-agent swarm-intelligence software architecture. Moroni and Manzolli described a system that uses mobile robots and video tracking along with evolutionary computation to compose and present sound art.
A significant topic in the application of evolutionary systems in the arts is computational aesthetic evaluation. Evolution proceeds by "survival of the fittest" and most engineering applications have objective measures of fitness that can allow a genetically based system to run generation after generation without human intervention. Applications in the arts, however, usually have an artist in the loop because aesthetic evaluation by computer remains an unsolved problem. Several papers at EvoMUSART touched on this problem including my own. I will summarize some of those ideas here in a future blog entry.
