Chapter 12 A neural-network approach to adaptive similarity and stimulus representations in cortico-hippocampal function

Abstract

We previously presented a top-down, connectionist model of cortico-hippocampal interaction during classical conditioning (Gluck & Myers, 1993) which argues that the hippocampal region adapts stimulus similarity according to two basic principles: a bias to differentiate—or reduce similarity—between stimuli that are predictive of different salient future events, and a bias to increase similarity between stimuli that co-occur and/or predict similar future events. The original cortico-hippocampal model of Gluck and Myers (1993) addressed a wide range of intact and hippocampal-lesion data. It also generated a number of novel and testable predictions about behavioral phenomena that should be hippocampal dependent. In more recent work (Myers, Gluck, & Granger, 1995) we have shown how at least one aspect of the proposed hippocampal - region function could arise from the substrate of the entorhinal cortex, the major sensory input to the hippocampal formation. Other work has shown how the role of cholinergic modulation from the medial septum can be incorporated into the model and relevant behavioral data explained (Myers, Ermita, Harris, Hasselmo, Solomon, & Gluck, 1996). This modeling suggests instead that the hippocampus is always active during normal learning. What should differ consistently between intact and hippocampal-lesioned animals are the strategies or methods by which they master tasks. This viewpoint suggests that attempts to differentiate and identify "hippocampal-dependent" or "hippocampal-independent" tasks may not be the most useful way to proceed. Rather, empirical emphasis should focus on tasks that can be solved by both intact and hippocampal-lesioned animals, and concentrate on ways in which generalization or transfer performance differs between the two groups.