New year, new scholarships

It is a great honor for Lukas to be funded by the oldest and largest organization for the promotion of talented young people in Germany. He is looking forward to meeting a network of promising and interesting people, and he hopes that valuable ideas and perhaps friendships will develop through mutual exchange. Lukas is a doctoral student in Jürgen Jost's group.

He plans to use the fellowship to investigate the mathematical formalization of concept emergence over the next two years. This is relevant in philosophy, linguistics, formal logic, psychology, neuroscience, and computer science, especially in the field of machine learning. Lukas attempts to approach the subject constructively from the logical-algorithmic realm. His starting point is the question of how one could formally describe a system (e.g., a recursive neural network or generalizations thereof) that, by interacting with a flow of information at the microscopic level (e.g., information in the form of pixels or photons), can gradually autonomously extract relationships between macroscopically relevant information (e.g., objects, motion sequences, etc.) that have meaning for the system's goals.

His research builds, among other things, on works in the field of “predictive coding”, which can perhaps be understood as a modern development of the principle of “unconscious inference” formulated as early as 1867 by Hermann von Helmholtz. According to this principle, an internal replication of the relationships of incoming information takes place by continuously (and “unconsciously”) verifying or discarding certain predictions about the flow of information and its effects in the system through a learning mechanism in the system. Lukas devotes particular attention to the question of what the system architecture and learning algorithm might look like in order to produce units of information in the system that could at least be called precursors of what we call concepts.