Hi all,
Tomorrow Eduardo González Sánchez will tell us about his master thesis on "Extracting physical parameters from an environment using AI-agents". See below for the abstract. We start at 3pm in zoom: https://ethz.zoom.us/j/362994444.
Best,
-joe
Abstract: An important challenge for the automation of science is to minimize the prior human knowledge built into the machine learning systems. A step in this direction was done in Phys. Rev. Lett. 124, 010508 (2020) where the relevant physical parameters were extracted from experimental data without using prior knowledge about the specific physical system. Here, we go one step further in minimizing prior knowledge: We do not consider the experimental data as given but train AI agents that learn to perform the experiments that provide the necessary data to extract the relevant parameters. To do so, we combine in a modular architecture techniques from reinforcement learning and deep learning. We demonstrate the working of our architecture with some toy examples. Reading out the parameters for the given examples is left for future work.