At the age of 33, I am a successful scientist in two fields: machine learning and psychology. Next to my Ph.D. in biological psychology, I trained myself in machine learning techniques, leading to a postdoc in this field. During research stays at Stanford University and UC San Diego, I was able to combine both areas of expertise. Currently, I am working as an AI Architect at Mercedes-Benz AG, developing the Mercedes Virtual Assistant.
I am interested in various topics including
- Investigating psychological research questions with machine learning
- Investigating state-of-the-art machine learning models with tools from psychology
- Large language models and diffusion models
- Efficient learning in generative, recurrent neural networks
- Artificial theory of mind
- Cognitive biases for machines
- Neuropsychological research on empathy for pain
My machine learning research has profited from my psychological background in various ways:
Most recently, I borrowed psychological tests, like the Cognitive Reflection Test, to investigate machine intuition in large language models like GPT-3. Since machine learning models are as obscure as humans’ mental states, psychological methods as well as the empirical rigor of hypothesis testing can help us shed light onto their inner working mechanisms. Machine learning research can also benefit from inspiration from human cognition. I solved the renowned Omniglot challenge developed by Lake, Salakuthinov, and Tenenbaum by including compositional encodings into a recurrent neural network. This was inspired by children who can easily generalize acquired knowledge by decomposing newly learned objects into parts and recombining them in new ways.
Not only machine learning research can benefit from psychology but also the other way round. I was investigating the racial bias in pain recognition with a computer vision model that detects the activation of different facial action units to determine whether specific perceptual properties, like shape or color of facial features, facilitate the detection of painful expressions.
I love challenging myself and pushing my limits. So, I couldn’t find a better sport than mountainbiking. Arguably, it is one of the toughest sports combining multiple requirements regarding endurance, technique, and courage to climb the steepest hills and send the sketchiest trails, drops, and jumps. Coming from a road bike background, mountainbiking demands this extra grit that I enjoy during my daily training rides.
Being an actress in musical performances was lots of fun but also very challenging for me. My strengths lie more in the academic than in the artistic field and I had not had any professional education in singing or dancing before. Nevertheless, I enjoyed making these new experiences very much.
I started a nationwide campaign with 200 people that aimed at promoting a web page that educates people to live a more environmentally friendly life. Tracking the web page views allowed to analyse the efficiency of different methods to promote the page.