Skip to main content


  • Poster presentation
  • Open Access

Characterising the performance of balanced memory networks

  • 1Email author,
  • 1,
  • 1,
  • 1 and
  • 1
BMC Neuroscience201314 (Suppl 1) :P88

  • Published:


  • Firing Rate
  • Associative Memory
  • Cortical Network
  • Excitatory Synapse
  • Hebbian Learning

In previous work [1], we investigated the associative memory performance of networks of Izhikevich neurons with various synaptic plasticity regimes. However, the firing rates observed were far higher than those observed in vivo. In their recent Science paper [2], Vogels et al. describe a model of associative memory that exhibits biologically plausible firing rates, using a network of Integrate and Fire (IAF) neurons in which the inhibitory to excitatory synapses are plastic. Their self-organising learning rule provides a homeostatic function, leading to balanced excitation and inhibition. Further, by achieving a globally balanced state, the network displays asynchronous irregular dynamics. This sparse pattern of activity, which is present in cortical networks in vivo, enables rapid responses to small changes in the input [2].

The patterns are stored via a simplified form of one-shot Hebbian learning of synapses between excitatory neurons. The plastic inhibitory to excitatory synapses serve to balance the excitation in the memory assembly, mirroring the potentiated excitatory synapses and thereby allowing the stored pattern to be suppressed when not activated by external stimuli. This is a feature lacking in the ANN attractor networks we have studied previously [1], where the activation of a pattern causes the network to reach a fixed point.

In this work, we characterise the performance of the Vogels' network using a metric, Effective Capacity, adapted from [3]. Vogels uses a random architecture, with each neuron having a 0.02 probability of being connected to any other neuron. However, analyses of the connectivity of cortical networks have suggested that they may have non-random features [4]. Having measured the performance of the network with random connectivity, we measure the effect on performance of a small world architecture [3]. These results are contrasted with a similar study of connectivity patterns in (non-spiking) ANN networks [5].

Authors’ Affiliations

Science and Technology Research Institute, University of Hertfordshire, Hatfield, Hertfordshire, AL10 9AB, UK


  1. Metaxas A, Maex R, Adams R, Steuber V, Davey N: Determinants of associative memory performance in spiking and non-spiking neural networks with different synaptic plasticity regimes. BMC Neuroscience. 2012, 13 (Suppl 1): P156-PubMed CentralView ArticleGoogle Scholar
  2. Vogels TP, Sprekeler H, Zenke F, Clopath C, Gerstner W: Inhibitory Plasticity Balances Excitation and Inhibition in Sensory Pathways and Memory Networks. Science. 2011, 334: 1569-1573. 10.1126/science.1211095.View ArticlePubMedGoogle Scholar
  3. Davey N, Calcraft L, Adams R: High capacity, small world associative memory models. Connection Science. 2006, 18: 247-264. 10.1080/09540090600639339.View ArticleGoogle Scholar
  4. Song S, Sjöström PJ, Reigl M, Nelson S, Chklovskii DB: Highly Nonrandom Features of Synaptic Connectivity in Local Cortical Circuits. PLoS Biol. 2005, 3: e68-10.1371/journal.pbio.0030068.PubMed CentralView ArticlePubMedGoogle Scholar
  5. Chen W, Maex R, Steuber V, Davey N: Clustering predicts memory performance in networks of spiking and non-spiking neurons. Front Comput Neurosci. 2011, 5: 14-PubMed CentralPubMedGoogle Scholar


© Metaxas et al; licensee BioMed Central Ltd. 2013

This article is published under license to BioMed Central Ltd. This is an Open Access article distributed under the terms of the Creative Commons Attribution License (, which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.