Learning recurrent dynamics in spiking networks

  1. Christopher M Kim  Is a corresponding author
  2. Carson C Chow  Is a corresponding author
  1. National Institutes of Health, United States

Abstract

Spiking activity of neurons engaged in learning and performing a task show complex spatiotemporal dynamics. While the output of recurrent network models can learn to perform various tasks, the possible range of recurrent dynamics that emerge after learning remains unknown. Here we show that modifying the recurrent connectivity with a recursive least squares algorithm provides sufficient flexibility for synaptic and spiking rate dynamics of spiking networks to produce a wide range of spatiotemporal activity. We apply the training method to learn arbitrary firing patterns, stabilize irregular spiking activity in a network of excitatory and inhibitory neurons respecting Dale's law, and reproduce the heterogeneous spiking rate patterns of cortical neurons engaged in motor planning and movement. We identify sufficient conditions for successful learning, characterize two types of learning errors, and assess the network capacity. Our findings show that synaptically-coupled recurrent spiking networks possess a vast computational capability that can support the diverse activity patterns in the brain.

Data availability

Example computer code that trains recurrent spiking networks is available at http://github.com/chrismkkim/SpikeLearning

Article and author information

Author details

  1. Christopher M Kim

    National Institute of Diabetes and Kidney and Digestive Diseases, National Institutes of Health, Bethesda, United States
    For correspondence
    chrismkkim@gmail.com
    Competing interests
    The authors declare that no competing interests exist.
    ORCID icon "This ORCID iD identifies the author of this article:" 0000-0002-1322-6207
  2. Carson C Chow

    National Institute of Diabetes and Kidney and Digestive Diseases, National Institutes of Health, Bethesda, United States
    For correspondence
    carsonc@niddk.nih.gov
    Competing interests
    The authors declare that no competing interests exist.
    ORCID icon "This ORCID iD identifies the author of this article:" 0000-0003-1463-9553

Funding

National Institute of Diabetes and Digestive and Kidney Diseases (Intramural Research Program)

  • Christopher M Kim
  • Carson C Chow

The funders had no role in study design, data collection and interpretation, or the decision to submit the work for publication.

Reviewing Editor

  1. Peter Latham, University College London, United Kingdom

Publication history

  1. Received: March 29, 2018
  2. Accepted: September 14, 2018
  3. Accepted Manuscript published: September 20, 2018 (version 1)
  4. Version of Record published: October 15, 2018 (version 2)

Copyright

This is an open-access article, free of all copyright, and may be freely reproduced, distributed, transmitted, modified, built upon, or otherwise used by anyone for any lawful purpose. The work is made available under the Creative Commons CC0 public domain dedication.

Metrics

  • 3,322
    Page views
  • 594
    Downloads
  • 24
    Citations

Article citation count generated by polling the highest count across the following sources: Scopus, Crossref, PubMed Central.

Download links

A two-part list of links to download the article, or parts of the article, in various formats.

Downloads (link to download the article as PDF)

Open citations (links to open the citations from this article in various online reference manager services)

Cite this article (links to download the citations from this article in formats compatible with various reference manager tools)

  1. Christopher M Kim
  2. Carson C Chow
(2018)
Learning recurrent dynamics in spiking networks
eLife 7:e37124.
https://doi.org/10.7554/eLife.37124

Further reading

    1. Computational and Systems Biology
    Shunpei Yamauchi, Takashi Nozoe ... Yuichi Wakamoto
    Research Article

    Intracellular states probed by gene expression profiles and metabolic activities are intrinsically noisy, causing phenotypic variations among cellular lineages. Understanding the adaptive and evolutionary roles of such variations requires clarifying their linkage to population growth rates. Extending a cell lineage statistics framework, here we show that a population’s growth rate can be expanded by the cumulants of a fitness landscape that characterize how fitness distributes in a population. The expansion enables quantifying the contribution of each cumulant, such as variance and skewness, to population growth. We introduce a function that contains all the essential information of cell lineage statistics, including mean lineage fitness and selection strength. We reveal a relation between fitness heterogeneity and population growth rate response to perturbation. We apply the framework to experimental cell lineage data from bacteria to mammalian cells, revealing distinct levels of growth rate gain from fitness heterogeneity across environments and organisms. Furthermore, third or higher order cumulants’ contributions are negligible under constant growth conditions but could be significant in regrowing processes from growth-arrested conditions. We identify cellular populations in which selection leads to an increase of fitness variance among lineages in retrospective statistics compared to chronological statistics. The framework assumes no particular growth models or environmental conditions, and is thus applicable to various biological phenomena for which phenotypic heterogeneity and cellular proliferation are important.

    1. Computational and Systems Biology
    Jeffrey Molendijk, Ronnie Blazev ... Benjamin L Parker
    Research Article

    Improving muscle function has great potential to improve the quality of life. To identify novel regulators of skeletal muscle metabolism and function, we performed a proteomic analysis of gastrocnemius muscle from 73 genetically distinct inbred mouse strains, and integrated the data with previously acquired genomics and >300 molecular/phenotypic traits via quantitative trait loci mapping and correlation network analysis. These data identified thousands of associations between protein abundance and phenotypes and can be accessed online (https://muscle.coffeeprot.com/) to identify regulators of muscle function. We used this resource to prioritize targets for a functional genomic screen in human bioengineered skeletal muscle. This identified several negative regulators of muscle function including UFC1, an E2 ligase for protein UFMylation. We show UFMylation is up-regulated in a mouse model of amyotrophic lateral sclerosis, a disease that involves muscle atrophy. Furthermore, in vivo knockdown of UFMylation increased contraction force, implicating its role as a negative regulator of skeletal muscle function.