Dokument

Summary:
What are the general principles that drive human learning in different situations? I argue that much of human learning can be understood with just three principles. These are generalization, adaptation, and simplicity. To verify this conjecture, I introduce a modeling framework based on the same principles. This framework combines the idea of meta-learning -- also known as learning-to-learn -- with the minimum description length principle. The models that result from this framework capture many aspects of human learning across different domains, including decision-making, associative learning, function learning, multi-task learning, and reinforcement learning. In the context of decision-making, they explain why different heuristic decision-making strategies emerge and how appropriate strategies are selected. The same models furthermore capture order effects found in associative learning, function learning and multi-task learning. In the reinforcement learning context, they resemble individual differences between human exploration strategies and explain empirical data better than any other strategy under consideration. The proposed modeling framework -- together with its accompanying empirical evidence -- may therefore be viewed as a first step towards the identification of a minimal set of principles from which all human behavior derives.

Das Dokument ist im Internet frei zugänglich - Hinweise zu den Nutzungsrechten