Székely, Anna (2026.05.15 - 2026.11.1 5)

Abstract: Understanding computational mechanisms that support transfer learning is a current topic of interest in both artificial intelligence and cognitive science research. In our research, conducted in collaboration with researchers at the University of Oxford, we aim to better understand the computations that support transfer learning in natural agents by describing human learning patterns. For this endeavor, we have access to a large-scale database of human behavior that was not typical in previous cognitive science experiments. To understand learning characteristics, we employ a “knowledge tracing” approach, through which we aim to identify the individual learning traits capable of predicting the performance of individual learners during the task transfer phase. To this end, we are developing hierarchical transformer models capable of learning the characteristics of individual tasks and individual learning traits, as well as predicting future performance. We are convinced that this research will enrich the international literature with valuable new insights regarding the understanding of transfer learning, and that the results will also serve as useful guidelines for educational development and the construction of personalized tutoring systems.