Liesbeth joined Volkswagen AI Research after her nascent interest in machine learning was further kindled by a presentation about empowerment, an information theoretic quantity that expresses the perceptible influence an agent has on his environment. She wanted to learn everything about it, and that is exactly what her bachelor thesis at the lab allows her to do. Her topic consists of giving an overview of modern ways of approximating empowerment, as computing it with classical methods is unfortunately intractable in most settings.
- intrinsically motivated reinforcement learning
- policy gradient methods
- reproducibility standards for deep reinforcement learning research