Abstract:

This chapter presents an emerging "case-based" system for decision making that centers on the retrieval of similar past episodes. It can be mathematically described by the economic model of case-based decision theory accompanied by the computational background of case-based reasoning. This system stands between model-free and model-based systems on the bias-variance tradeoff and holds advantages over both when faced with novel and complex problems. The hippocampus and related medial temporal lobe structures provide a neural substrate for case-based calculations by instantiating stimulus and reward associations. This neurocomputational understanding sheds light on how such calculations can also contribute to or interfere with model-free and model-based control.