Improving Recall in an Associative Neural
Network Model of the Hippocampus
Nikolaos Andreakos 1 , Shigang Yue 1 , and Vassilis Cutsuridis 1,2(B)
1 School of Computer Science, University of Lincoln, Lincoln, UK
{nandreakos,syue,vcutsuridis}@lincoln.ac.uk
2 Lincoln Sleep Research Center, University of Lincoln, Lincoln, UK
Abstract. The mammalian hippocampus is involved in auto-association and
hetero-association of declarative memories. We employed a bio-inspired neural
model of hippocampal CA1 region to systematically evaluate its mean recall quality against different number of stored patterns, overlaps and active cells per pattern.
Model consisted of excitatory (pyramidal cells) and four types of inhibitory cells:
axo-axonic, basket, bistratified, and oriens lacunosum-moleculare cells. Cells were
simplified compartmental models with complex ion channel dynamics. Cells’ firing was timed to a theta oscillation paced by two distinct neuronal populations
exhibiting highly regular bursting activity, one tightly coupled to the trough and
the other to the peak of theta. During recall excitatory input to network excitatory
cells provided context and timing information for retrieval of previously stored
memory patterns. Dendritic inhibition acted as a non-specific global threshold
machine that removed spurious activity during recall. Simulations showed recall
quality improved when the network’s memory capacity increased as the number of
active cells per pattern decreased. Furthermore, increased firing rate of a presynaptic inhibitory threshold machine inhibiting a network of postsynaptic excitatory
cells has a better success at removing spurious activity at the network level and
improving recall quality than increased synaptic efficacy of the same threshold
machine on the same network of excitatory cells, while keeping its firing rate
fixed.
Keywords: Associative memories · Brain · Inhibition
1 Introduction
Associative memory (AM) is the ability to learn and remember the relationship between
items, events, places and/or objects which may be unrelated [18]. AM is one of the oldest
artificial neural networks’ paradigms [19, 20]. In these models storing patterns was done
via changes in connection strengths between artificial neurons which crudely mimicked
biological ones. Old memories were recalled when a noisy, partial or complete version
of a previously stored pattern was presented to the network. However, these AM devices
were not very flexible. They had to be told when to store a memory pattern and when to
recall it.
© Springer Nature Switzerland AG 2020
V. Vouloutsi et al. (Eds.): Living Machines 2020, LNAI 12413, pp. 1–11, 2020.
https://doi.org/10.1007/978-3-030-64313-3_1
Network Model of the Hippocampus
Nikolaos Andreakos 1 , Shigang Yue 1 , and Vassilis Cutsuridis 1,2(B)
1 School of Computer Science, University of Lincoln, Lincoln, UK
{nandreakos,syue,vcutsuridis}@lincoln.ac.uk
2 Lincoln Sleep Research Center, University of Lincoln, Lincoln, UK
Abstract. The mammalian hippocampus is involved in auto-association and
hetero-association of declarative memories. We employed a bio-inspired neural
model of hippocampal CA1 region to systematically evaluate its mean recall quality against different number of stored patterns, overlaps and active cells per pattern.
Model consisted of excitatory (pyramidal cells) and four types of inhibitory cells:
axo-axonic, basket, bistratified, and oriens lacunosum-moleculare cells. Cells were
simplified compartmental models with complex ion channel dynamics. Cells’ firing was timed to a theta oscillation paced by two distinct neuronal populations
exhibiting highly regular bursting activity, one tightly coupled to the trough and
the other to the peak of theta. During recall excitatory input to network excitatory
cells provided context and timing information for retrieval of previously stored
memory patterns. Dendritic inhibition acted as a non-specific global threshold
machine that removed spurious activity during recall. Simulations showed recall
quality improved when the network’s memory capacity increased as the number of
active cells per pattern decreased. Furthermore, increased firing rate of a presynaptic inhibitory threshold machine inhibiting a network of postsynaptic excitatory
cells has a better success at removing spurious activity at the network level and
improving recall quality than increased synaptic efficacy of the same threshold
machine on the same network of excitatory cells, while keeping its firing rate
fixed.
Keywords: Associative memories · Brain · Inhibition
1 Introduction
Associative memory (AM) is the ability to learn and remember the relationship between
items, events, places and/or objects which may be unrelated [18]. AM is one of the oldest
artificial neural networks’ paradigms [19, 20]. In these models storing patterns was done
via changes in connection strengths between artificial neurons which crudely mimicked
biological ones. Old memories were recalled when a noisy, partial or complete version
of a previously stored pattern was presented to the network. However, these AM devices
were not very flexible. They had to be told when to store a memory pattern and when to
recall it.
© Springer Nature Switzerland AG 2020
V. Vouloutsi et al. (Eds.): Living Machines 2020, LNAI 12413, pp. 1–11, 2020.
https://doi.org/10.1007/978-3-030-64313-3_1