Between memory systems analysis and synthesis: Biologically plausible and mechanistically interesting aspects of declarative memory modelling across brain areas and time
CBN (Computational Biology and Neurocomputing) seminars
Friday 16 May 2014
to 11:00 at
Florian Fiebig (CB/CSC/KTH and Institute for Adaptive and Neural Computation, University of Edinburgh)
This seminar outlines a scientific plan to a PhD Thesis. It covers both what I have done so far and what I am aiming for while noting the motivations and reasoning behind certain modelling ideas in memory systems modelling. As such it might also be good overall introduction for anyone interested in memory systems modelling.
I will outline of our current understanding of the human declarative memory system and will address the presumed roles of several brain structures involved in memory consolidation (pre-frontal cortex, hippocampus and its various substructures, as well as neocortex) while focusing specifically on the temporal aspect: How can we learn and forget on the scale of seconds while retaining some memories after decades? What are the presumed mechanisms for the consolidation of memory and how can we model these? What is the difference between recall and familiarity in memory retrieval? How do we think about the neural coding of memory and the role of schemas in memory consolidation? And what does all of this have to do with applied memory techniques and learning strategies we use as academics all the time?
I will present my modelling work on a three-stage complementary learning system with an intrinsic mechanism for autonomous consolidation using Bayesian Confidence Propagation Neural Networks. Memory consolidation in such a model can be favorably compared to experimental data via simulated lesioning experiments and memory modulating agents (such as Benzodiazepines, Ethanol or other plasticity regulators), as well as observations of memory reactivations during slow-wave-sleep. From there we will explore ideas for future improvement, functional validation, implementation on neuromorphic hardware, robotic applications and hopefully have enough time for a stimulating discussion afterwards.