Inference Engines
for Darius Jones, Michael Nicolella, Jordan Voelker, and Bonnie Whiting
The title refers to an inference engine as a neural network component both in metaphorical and direct ways. The musical content consists of a guided improvisation session partially driven by the musicians’ inferences on the written material, reflecting the musician mind as a metaphorical inference engine. The score also provides an input for an actual inference engine, generating a fluid video-notation that guides the improvisation.
These elements come together within a context that focuses on racially-biased practices in facial recognition applications. Weights from a biased and insufficient dataset are fed into a third inferencing engine that is optimized for real-time audio processing to achieve a distortion effect. A deliberate distancing between the musical process and the sound effect serves to emphasize the essential identity of the distortion as an output of a faulty neural network. Its separate and insistent presence reminds us the very real consequences of using such networks.