Theodora Drăgan points to a robot arm in one of the laboratories at the Fraunhofer Institute for Cognitive Systems (IKS) – teaching swarms of robots to organize themselves efficiently in the future would be a dream come true for her. Theodora is a research associate at the Fraunhofer IKS, where she conducts research on classical and quantum machine learning. Possible applications are always on her mind.

"My path found me, not the other way around," says Theodora about her beginnings in the field of quantum computing and describes herself as lucky. After completing her Bachelor's degree in computer science and engineering in Romania, the 26-year-old came to Germany four years ago to complete a Master's degree in computer science at the Technical University of Munich. Theodora was immediately impressed by the lecture on the introduction to quantum computing. From then on, she took every course possible in the field, which for Theodora was the perfect mix of computer science and physics. "I really loved the lectures and my professor was amazing!". This professor, Christian Mendl, ultimately also supervised her Master's thesis on quantum reinforcement learning, which Theodora wrote at Fraunhofer IKS. According to Theodora, it was the "fantastic guidance of Jeanette Lorenz" that significantly shaped her further work.

Theodora likes to explain quantum reinforcement learning using the example of a mouse in a maze that has to find the right way to get a piece of cheese. Reinforcement learning is about the idea that an agent, a computer program that is independent to a certain extent, learns through interaction to solve a specific environment problem, in this example the maze. There is a reward for the correct or best solution – in the case of the mouse, this is cheese. Theodora's mouse is her algorithm. But what does a reward for an algorithm look like? A function determines which actions are good or bad. "It's actually exactly the same as with the mouse," explains the scientist, "only here it's a number, nothing more and nothing less. If you perform well, that's your reward: a higher number," says Theodora with a smile. This can also be understood as the accuracy of a neural network. Neural networks are machine-learning programs inspired by the networks of biological neurons in the human brain. Theodora's work focuses on replacing the neural networks that process input from the environment to find the right decision with different quantum circuits. This is one side of quantum reinforcement learning. "On the other side, we try to exploit the noise and superposition that exists in quantum circuits to mimic the uncertainty in human decisions, for example," explains Theodora.

Finding faster solutions with fewer interactions

Theodora discovered her interest in how things work on the inside at an early age. Her father, a network engineer, liked to bring computers home and take them apart while Theodora watched. "I found it quite fascinating," says the 26-year-old, "so as a child, I wanted to become an IT engineer like my father." Her work now focuses less on hardware, but her original choice of study in engineering stems from this. "I never specifically dreamed of becoming a scientist, but I always dreamed of finding answers to difficult questions," says Theodora, describing her subsequent path. This is exactly what fascinates her about quantum computing: The novelty and the opportunity to explore new avenues – while regularly questioning what she thinks she knows about physics and computer science.

When Theodora works on algorithms, however, it is not so important whether she writes a classical or a quantum algorithm, as the level of abstraction is so high anyway. "You can also easily integrate your quantum circuits into classical reinforcement learning environments. That's not a problem," she explains. The advantage of quantum circuits, as Theodora hopes, is that an agent requires fewer interactions with the environment to learn, because such interactions can be expensive: "Especially when we think of medical applications, the agent cannot interact with the environment over and over again, because the environment is the patient." In a paper that Theodora and her colleagues published last year, they were able to show that quantum reinforcement learning performed better than its classical counterpart in a path planning problem. "And significantly better," emphasizes Theodora, visibly pleased. "I could clearly see that the quantum solution learned faster." What drives Theodora in her research is the pursuit of excellence, which she also observes and admires in Jeanette Lorenz. "It motivates me immensely to be very confident in my research, to have confidence in what I'm doing," she says. In return, the scientist is willing to sit at her desk until late at night to solve a complicated problem. The more effort she puts into a project, the prouder she is of the results in the end.