Figure - PMC

Skip to main content

An official website of the United States government

Here's how you know

Here's how you know

Official websites use .gov
A .gov website belongs to an official government organization in the United States.

Secure .gov websites use HTTPS
A lock ( ) or https:// means you've safely connected to the .gov website. Share sensitive information only on official, secure websites.

View full-text article in PMC

. 2013 Oct 24;9(10):e1003272. doi: 10.1371/journal.pcbi.1003272

Search in PMC
Search in PubMed
View in NLM Catalog
Add to search

© 2013 Yger, Harris

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 author and source are properly credited.

PMC Copyright notice

(A) Network illustration. A set of 3600 excitatory and 900 inhibitory recurrently connected neurons are driven by an external excitatory input drawn from a cochleogram simulated as before. Excitatory synapses within the network are plastic while inhibitory synapses and external inputs are kept fixed. A population of linear readout neurons use the spike counts of the recurrent excitatory neurons to classify the spoken digits. (B) Illustration of population activity in the network, before and after learning, in response to a particular digit. The rasters show activity of the entire population of excitatory and inhibitory neurons (red and blue) to a single digit presentation; the lower curves show population-averaged firing rate throughout this trial. Note that training produces no visible change in global network dynamics, which maintains an asynchronous regular state. (C–E) Distributions of ISI CVs, firing rates, and pairwise correlation coefficients (averaged over 2000 randomly chosen pairs of cells) in the network before and after learning with rate constraint only, rcSTDP, or with the Convallis learning rule. Note that none of the learning rules produce a change in any of these measures of network dynamics. Error bars show the standard deviation. (F) Distribution of membrane potential skewness for 200 randomly chosen cells in the network before or after learning. Note that skewness is highest with the Convallis rule. (G) Distribution of the tuning sharpness (as measured by F-statistic) for all neurons before and after learning. Inset displays the mean of the distributions. Error bars show standard deviation. (H) Classification performance as a function of the number of neurons considered by the external classifier. Errors bars show s.e.m over 10 different simulations, run independently from different random seeds. While Convallis learning produces improved performance, rate constraint did not, and rcSTDP produced a smaller but still significant improvement ().

Inline graphic — (A) Network illustration. A set of 3600 excitatory and 900 inhibitory recurrently connected neurons are driven by an external excitatory input drawn from a cochleogram simulated as before. Excitatory synapses within the network are plastic while inhibitory synapses and external inputs are kept fixed. A population of linear readout neurons use the spike counts of the recurrent excitatory neurons to classify the spoken digits. (B) Illustration of population activity in the network, before and after learning, in response to a particular digit. The rasters show activity of the entire population of excitatory and inhibitory neurons (red and blue) to a single digit presentation; the lower curves show population-averaged firing rate throughout this trial. Note that training produces no visible change in global network dynamics, which maintains an asynchronous regular state. (C–E) Distributions of ISI CVs, firing rates, and pairwise correlation coefficients (averaged over 2000 randomly chosen pairs of cells) in the network before and after learning with rate constraint only, rcSTDP, or with the Convallis learning rule. Note that none of the learning rules produce a change in any of these measures of network dynamics. Error bars show the standard deviation. (F) Distribution of membrane potential skewness for 200 randomly chosen cells in the network before or after learning. Note that skewness is highest with the Convallis rule. (G) Distribution of the tuning sharpness (as measured by F-statistic) for all neurons before and after learning. Inset displays the mean of the distributions. Error bars show standard deviation. (H) Classification performance as a function of the number of neurons considered by the external classifier. Errors bars show s.e.m over 10 different simulations, run independently from different random seeds. While Convallis learning produces improved performance, rate constraint did not, and rcSTDP produced a smaller but still significant improvement ().