[Next] [Up] [Previous] Complementarity, Memory Consolidation and Context Integration in the Medial Temporal Lobe Anders Sandberg Goal To study how the medial temporal lobe (MTL) and the cerebral cortex interacts, especially conceptual models for complementary learning systems and how they may support memory consolidation and context integration through ``fast'' and ``slow'' learning. Background The hippocampus and the surrounding medial temporal lobe system (the parahippocampal, perirhinal and entorhinal cortices [ZMS93]) is crucial to memory consolidation, the process where declarative memories are gradually encoded into the neocortex [SA95] The MTL appears to be active during memory consolidation and learning declarative memories [PEI], while parts of the neocortex becomes active during recall (cf. [Fus94]). It is at present not clear what role the MTL plays in the consolidation and reinstatement processes, and there has been made many suggestions about how the system works. It has been suggested that the hippocampus plays role as spatial memory [ON78]. This is supported by the existence of place cells in the hippocampus which fire when the animal is in a certain location but is contradicted in humans by the fact that lesions to the hippocampus also causes deficits in non-spatial memory. One alternative model is that the hippocampus binds together different representations into a ``chunk'' which is then learned by the neocortex [Squ92, ]; the place cells are in this model context cells [MG97]. A different group of models is that the hippocampus acts as intermediate storage of memory before it is transferred into long-term memory in the neocortex, an idea originally proposed by Marr [Mar71]. This could be done using a rehearsal mechanism that strengthens the connections between different cortical areas corresponding to the same original activation pattern [Mis90]. The MTL only needs to store an ``index'' pattern which allows the retrieval of the original pattern of cortical activation. This is similar to the pseudorehearsal model of Robins [Rob96]. There exists large differences in learning speed between neocortical memory and hippocampal memory, as is evidenced by the need for repetition in procedural learning and semantic learning rates in amnesiacs contrasted by the fact that episodic memory in general can be encoded with a single experience. Peter Milner has suggested that the hippocampal synapses are ``soft'' and capable of changing quickly, both in learning and forgetting, while neocortical synapses are ``hard'' and change slowly in response to new stimuli and weaken slowly. The hippocampus can reactivate the cortical representations during recall, and through repeated activations create hard links that remain as long term memories long after the hippocampal representations have faded [Mil89, ]. This complementary model of necortical-MTL interactions fits together with the conceptual models of McClelland et al. [MMO95] where the difference in plasticity is explained as a solution to the stability-plasticity dilemma and the catastrophic interference effect (where new learning interferes with recall of older memories) which is observed in many simple memory models but not in humans [Rat90] (it should be noted that this effect does not necessarily occur in all models). There interference can be avoided within this framework, for example through rehearsal and pseudorehearsal which would correspond to the MTL storing information which is then gradually replayed to train the neocortex, possibly during sleep [Rob96] or together with ongoing experience [MG97]. So far most models of the MTL has been purely conceptual and with minimal ties to biological relevancy or concentrated on low-level properties of the hippocampal network. This project intends to study conceptual models of MTL and its interactions with neocortex with an eye towards biological reasonability, and to test the viability of the complementarity hypothesis. Another important area in memory consolidation is how it is influenced by attention, the relevance of information and the emotional state of the animal. It is known that various hormones influence memory, likely by influencing modulatory systems which in turn regulate the strength of associations [MSW91]. The MTL seems ideally placed for influencing associative strengths due to its limbic and mesencephalic connections which may influence consolidation and the internal plasticity. It is hence natural to keep an eye open for how this form of modulation can be accommodated in the complementary model. Time plan with rough time estimates Literature survey (4 weeks) * Theories of MTL function * Different time constants for plasticity and complementarity * Hopfield networks with clipped weights [NTCD86] * Relevancy signals Presentation as a seminar and/or essay. Consolidation Models (12 Weeks) Develop a model for MTL-neocortical interaction and consolidation with different learning time constants. * Develop an environment and MTL-NCX model to work with. * Test how well networks with short time constants can support memory consolidation in long time constants. How does their size and structure affect the performance? * Test different training models, preferably biologically reasonable. * Compare with results and models in the literature. Presentation as an article. Context (4 Weeks) * Study self-organized context-binding, abstraction and time integration of experiences. * Are place-cells due to self-organization in a general context-binding system, or are they biologically determined? * Memory modulation using relevance-signals and hormones. * Source memory, episodic and semantic memory: can they be explained in the model? Presentation as an article. Possibly related subjects Sleep, limbic system, cholinergic pathways, frontal lobe memory systems. References AS94 Pablo Alvarez and Larry R. Squire. Memory consolidation and the medial temporal lobe: A simple network model. Proc. Natl. Acad. Sci. USA, 91:7041-7045, July 1994. BWP95 Andrea Bibbig, Thomas Wennekers, and Günther Palm. A neural network model of the cortico-hippocampal interplay and the representation of contexts. Behavioural Brain Research, 66:169-175, 1995. Fus94 Joaquín M. Fuster. Memory in the Cerebral Cortex. MIT Press, 1994. Mar71 D. Marr. Simple memory: a theory for the archicortex. Philosophical Transactions of the Royal Society of London, B, 262:23-81, 1971. MG97 James L. McClelland and Nigel H. Goddard. Considerations arising from a complementary learning systems perspective on hippocampus and neocortex. Hippocampus, 6:654-665, 1997. Mil89 P. Milner. Neuropsychologia, 1989. Mis90 M. Mishkin. Vision, Memory and the Temporal Lobe, pages 7-15. Elsevier, NY, 1990. MMO95 J.L McClelland, B.L. McNaughton, and R. C. O'Reilly. Why there are complementary learning systems in the hippocampus and neocrotex: Insights from the success and failures of connectionist models of learning and memory. Psychological Review, 102:419-457, 1995. MSW91 Joe L. Martinez, Gery Schulteis, and Susan B. Weinberger. Learning and Memory: A Biological View, chapter 4, pages 149-198. Academic Press, second edition, 1991. NTCD86 J.P. Nadal, G. Toulouse, J.P. Changeux, and S. Dehaene. Networks of formal neurons and memory palipsests. Europhysics Letters, 1(10):535-542, May 1986. ON78 J. O'Keefe and L. Nadel. The Hippocampus as a Cognitive Map. Clarendon Press, Oxford, 1978. PEI Karl Magnus Petersson, Christina Elfgren, and Martin Ingvar. A dynamical role of the medial temporal lobe during retrieval of declarative memory in man. Unpublished article. Rat90 R. Ratcliff. Connectionist models of recognition memory: Constraints imposed by learning and forgetting functions. Psychol. Rev., 97:285-308, 1990. Rob96 Anthony Robins. Consolidation in neural networks and in the sleeping brain. Connection Science, 8(2):259-275, 1996. SA95 Larry R. Squire and Pablo Alvarez. Retogerade amnesia and memory consolidation: a neurobiologival perspective. Current Opinion in Neurobiology, 5:169-177, 1995. Squ92 L. R. Squire. Memory and the hippocampus: A synthesis from findings with rats, monkeys, and humans. Psychological Review, 99:195-231, 1992. ZMS93 S. Zola-Morgan and L. R. Squire. Neuroanatomy of memory. Annu. Rev. Neurosci., 16:547-563, 1993. About this document ... Complementarity, Memory Consolidation and Context Integration in the Medial Temporal Lobe This document was generated using the LaTeX2HTML translator Version 96.1-h (September 30, 1996) Copyright © 1993, 1994, 1995, 1996, Nikos Drakos, Computer Based Learning Unit, University of Leeds. The command line arguments were: latex2html -split 0 -ascii_mode project.tex. The translation was initiated by Anders Sandberg on Mon May 12 15:21:41 MET DST 1997 ---------------------------------------------------------------------------- [Next] [Up] [Previous] Anders Sandberg Mon May 12 15:21:41 MET DST 1997