Presentation on theme: "REMERGE: A new approach to the neural basis of generalization and memory-based inference Dharshan Kumaran, UCL Jay McClelland, Stanford University."— Presentation transcript:
REMERGE: A new approach to the neural basis of generalization and memory-based inference Dharshan Kumaran, UCL Jay McClelland, Stanford University
Proposed Architecture for the Organization of Semantic Memory McClelland, McNaughton & O’Reilly, 1995 color form motion action valance Temporal pole name Medial Temporal Lobe
Two Questions If extraction of generalizations depends on gradual learning, how do we form generalizations and inferences shortly after initial learning? Why do some studies find evidence consistent with the view that an intact MTL facilitates certain types of generalization in memory?
Relational Theory of Memory (Eichenbaum & Cohen) Proposes that elements of related memories become linked within the same memory trace, and that the formation of such linkages is a critical function of the MTL.
REMERGE: Recurrence and Episodic Memory Results in Generalization Holds that several MTL based item representations may work together through recurrent activation Draws on classic exemplar models (Medin & Shaffer, 1978; Nosofsky, 1984) Extends these models by allowing similarity between stored items to influence performance, independent of direct activation by the probe (McClelland, 1981) Demonstrates the strong dependence of some forms of generalization and inference on the strength of learning for trained items
Phenomena Considered Benchmark Simulations – Categorization – Recognition memory Acquired Equivalence Associative Chaining – In paired associate learning – In hippocampal reactivation during sleep Transitive Inference – Effects of increasing study – Effects of sleep
Associative Chaining Study: – AB, XY – BC, YZ Test: – A: B or Y – A: C or Z A B C X Y Z
Hippocampal Reactivation After Maze Exploration Replays in Remerge: Forward: 51% Backward:31% Crossed:18% Disjoint:<1%
Growth in Generalization with Increasing Premise Strength
Discussion As we’ve known for quite some time – Generalization and Inference can be supported by exemplar models Should we, then, simply abandon the complementary learning theory, and just make it exemplars all the way down? I think not – – Cortical learning supports changes in the ‘features’ that serve as the basis for exemplar learning – And clearly, retrograde amnesia studies support = an MTL basis for recent memory = a neocortical basis for remote memory A future challenge is to develop an fully integrated neuro-computational theory of memory integrating MTL and neocortical influences