A dynamic model of associative memory storage and recall

1
A dynamic model of associative memory storage and
recall

• Janette Clark
• Dept of Computer Science Maths
• University of Stirling

2
Project Proposal

• The aim of this project is to build a computer
  simulation of a dynamic model of memory storage
  and recall in the mammalian brain.
• Previous associative memory network models,
  including those that attempt to be biologically
  realistic, greatly simplify the inhibitory and
  neuromodulatory components of the network and
  treat storage and recall as distinct phases.
• I propose to explore a different theory of
  memory, that memory can be used in a dynamic mode
  in which it is continually receiving input
  patterns and recalling patterns.
• The model will be based on the associative
  memory theory of memory function implemented as
  an associative network.
• The network will be based on the architecture of
  the CA1 and CA3 regions of the mammalian
  hippocampus.

3
Computational Neuroscience
electrode

• We now have considerable data about the human
  brain its biological and chemical make up and
  electrical properties. However, how it carries
  out certain functions remains largely in the
  realms of hypothesis and experimental research.
• Computer simulated models are essential if we are
  to understand brain functionality, such as how we
  store and recall memory. Traditional research
  techniques are problematic within the field of
  neuroscience.
• it is not easy to experiment on live models and
  in vitro experimentation is subject to complex
  environmental influences

4
Brain Basics The Neuron

• Fundamental to brain function are the nerve cells
  called neurons.
• There are approximately 100 billion neurons in
  the brain, each of which connects, or synapses,
  to potentially thousands of other cells to form
  networks of neurons that are pathways through
  which information in the nervous system is
  transmitted or stored.
• There are somewhere in the region of 100 trillion
  synapses in the brain. Thats about 1 billion
  chemical connections per cubic mm of cortical
  grey matter!

Schematic of a neuron. Inset, representation of
the synaptic cleft
5
Learning Memory

• Synaptic plasticity the basis of learning and
  memory
• This is the ability of an organisms synapses to
  change in response to a stimulus.
• Hebbian learning.
• Theory of the biophysics of learning devised by
  psychologist Donald Hebb in 1949.
• When an axon of a cell A is near enough to
  excite cell B or repeatedly or persistently takes
  part in firing it, some growth or metabolic
  change takes place in both cells such that As
  efficiency, as one of the cells firing B, is
  increased.
• D Hebb. The Organization of Behaviour (1949)

6
Memory Models Associative Memory
Types of Associative Memory

• Theories of memory associative memory model
• We are often able to recall memories from small
  partial triggers.
• Essential for this type of memory is the ability
  to form associations.
• Synaptic plasticity is the necessary ingredient
  behind associative abilities in the brain.

7
Biology of Associative Memory

• If circuits in the nervous system act as
  associative memories, they must be able to
  dynamically interleave storage of new patterns
  with recall of old patterns.
• The cellular and circuit-level mechanisms that
  may provide this functionality have been
  postulated by Paulsen and Moser (TINS,
  21(7)273-278, 1998), with particular reference
  to areas CA3 and CA1 of the mammalian hippocampus.

Schematic of young rat hippocampus
8
Hippocampal Microcircuit

• Autoassociative memories rely on extensive
  recurrent connectivity between the neurons in the
  network. Sufficient recurrent connectivity is
  found in the CA3 region of the hippocampus.
• Any feedforward network can act as a
  heterassociative memory. The network formed by
  axons from the CA3 pyramidal cells synapsing onto
  CA1 pyramidal cells in the hippocampus has been
  postulated to act in this way.

A schematic representation of the connections of
the hippocampus. Rolls (1989)
9
A Dynamic Model of Memory Function Interneurons
Schematic diagram of some GABAergic cells in the
hippocampus. Principal cells (that is, granule
cells or pyramidal cells, shown in black) are
contacted by GABAergic interneurones via
feed-forward (left) as well as feedback circuits
(right). (From Paulsen and Moser 1998)
10
A Dynamic Model Brain Waves

• Two brain rhythm states of interest in my
  research are that of theta (5-8Hz) and gamma
  (30-50Hz).
• Lisman and Idiart (1995) proposed a model whereby
  multiple memories can be stored in a single
  neuronal network by recording, or storing, each
  memory in a different high frequency (gamma)
  subcycle of a low-frequency oscillation. Memory
  patterns then repeat on each low-frequency
  (theta) oscillation, reinforcing the neuronal
  pattern, strengthening the synapses related to
  this memory pattern on each repetition.

11
A Dynamic Model Resonance

• What determines the frequency range of each brain
  rhythm? (Hutcheon Yarom)
• Can we utilise the notion of frequency-preference
  in neurons?
• Hutcheon and Yarom postulate that there is a
  common element underlying the diverse mechanisms
  of frequency preference in neurons Resonance.
• Can we use the property of resonance as a high or
  low band pass filter? A switching mechanism for
  simultaneous, dynamic memory storage and recall?

12
The Model

• A CA1 pyramidal cell compartmental model has been
  built based on the previous work by Panayiota
  Poirazi and Dr Bruce Graham, Stirling
  University, using the Neuron simulation
  environment.
• Raw cell data of a young, in vitro, rat
  hippocampus was imported from the Duke /
  Southampton Archive of Neuronal Morphology.
• The entire model consists of 183 compartments and
  includes a variety of active and passive membrane
  mechanisms known to be present in CA1 pyramidal
  cells.

from Institute of Molecular Biology and
Biotechnology, Crete, Greece
13
The Simulator Interface
14
Proposed Network Model

• A CA3 cell will be developed based on data from
  Dr Stuart Cobb, Glasgow University.
• Thus allowing me to simulate the network
  connections and phenomenon thought to participate
  in declarative associative memory.
• The theory of dynamic associative memory using
  resonant behaviour will be applied.
• Models that allow simulation of the function of
  the hippocampal formation have broad application,
  in particular within the areas of research
  involved in trying to understand pathological
  neural disorders such as Alzheimers disease and
  epilepsy.
• To this end the project will be extended to
  incorporate the work of Manuel Sanchez-Montanes,
  University of Madrid, on epilepsy and it will be
  investigated whether we can simulate epileptic
  phenomenon in the model and thus postulate
  potential causal effects.

15
The End