Freiburg 2019 – wissenschaftliches Programm
FM 65.7: Poster
Mittwoch, 25. September 2019, 16:30–18:30, Tents
Autonomous learning agents make sense of a complex environment by proposing latent variables — •Katja Ried1, Benjamin Eva2, Thomas Müller2, and Hans J. Briegel1,2 — 1Institute for Theoretical Physics, University of Innsbruck — 2Department of Philosophy, University of Konstanz
Learning agents are becoming powerful tools that help us understand and control complex systems, both classical and quantum. One major challenge in their development is the exponentially large number of possible inputs they may encounter. Humans overcome this challenge by decomposing their perceptions, identifying features, variables and concepts. We present a minimal example of an artificial learning agent that, upon interacting with a structured environment, autonomously develops an internal representation that uses latent (unobserved) variables to organize its knowledge. We further show how this representation allows the agent to apply previous knowledge to situations it has not encountered before.