A schematic diagram depicting the recall of a sequence of memory items
when the network containing the pool of memory items is triggered by a
stimulus.
A model that shows how connections in the brain must change to form
memories could help to develop artificial cognitive computers
Exactly how memories are stored
and accessed in the brain is unclear. Neuroscientists, however, do know that a
primitive structure buried in the center of the brain, called the hippocampus,
is a pivotal region of memory formation. Here, changes in the strengths of
connections between neurons, which are called synapses, are the basis for
memory formation. Networks of neurons linking up in the hippocampus are likely
to encode specific memories.
Since direct tests cannot be
performed in the brain, experimental evidence for this process of memory
formation is difficult to obtain but mathematical and computational models can
provide insight. To this end, Eng Yeow Cheu and co-workers at the A*STAR
Institute for Infocomm Research, Singapore, have developed a model that sheds
light on the exact synaptic conditions required in memory formation1.
Their work builds on a previously
proposed model of auto-associative memory, a process whereby a memory is
retrieved or completed after partial activation of its constituent neural
network (see image). The earlier model proposed that neural networks encoding
short-term memories are activated at specific points during oscillations of
brain activity. Changes in the strengths of synapses, and therefore the
abilities of neurons in the network to activate each other, lead to an
auto-associative long-term memory.
Cheu and his team then adapted a
mathematical model that describes the activity of a single neuron to
incorporate specific characteristics of cells in the hippocampus, including
their inhibitory activity. This allowed them to model neural networks in the hippocampus
that encode short-term memories. They showed that for successful formation of
auto-associative memories, the strength of synapses needs to be within a
certain range: if synapses become too strong, the associated neurons are
activated at the wrong time and networks become muddled, destroying the
memories. If they are not strong enough, however, activation of some neurons in
the network is not enough to activate the rest, and memory retrieval fails.
As well as providing insight into
how memories may be stored and retrieved in the brain, Cheu thinks this work
also has practical applications. “This study has significant implications in
the construction of artificial cognitive computers in the future,” he says. “It
helps with developing artificial cognitive memory, in which memory sequences
can be retrieved by the presentation of a partial query.” According to Cheu, one can compare it to a
single image being used to retrieve a sequence of images from a video clip.
The A*STAR-affiliated researchers
contributing to this research are from the Institute for Infocomm Research
References
- Cheu, E. Y., Yu, J., Tan, C. H. & Tang, H.
Synaptic conditions for auto-associative memory storage and pattern
completion in Jensen et al.’s model of hippocampal area CA3. Journal
of Computational Neuroscience advance online publication, 30 May
2012 (doi: 10.1007/s10827-012-0394-8). | article
No comments:
Post a Comment