Erlangen 2018 – scientific programme
Parts | Days | Selection | Search | Updates | Downloads | Help
Q: Fachverband Quantenoptik und Photonik
Q 16: Quantum Information and Simulation
Q 16.3: Talk
Monday, March 5, 2018, 14:30–14:45, K 1.020
Projective simulation memory network for solving toy and complex problems — •Alexey Melnikov1, Vedran Dunjko2, Hendrik Poulsen Nautrup1, and Hans Briegel1,3 — 1Institute for Theoretical Physics, University of Innsbruck — 2Max-Planck-Institute for Quantum Optics — 3Department of Philosophy, University of Konstanz
The projective simulation (PS) model is a physical approach to artificial intelligence. In the PS model, learning is realized by internal modification of the episodic memory network, both in terms of its structure and the weights of its edges. Through interactions with a task environment, the PS memory network adjusts itself dynamically, so as to increase the probability of performing better in subsequent time steps. Here we consider several examples of environments, in which the PS agent does self-adjustments due to the glow, the generalization and the meta-learning mechanisms. The emphasis is made on examples of the PS agent applied to quantum optics experiments in which the agent autonomously learns to reach various entanglement classes.