Local online learning in recurrent networks with random feedback

  1. James M Murray  Is a corresponding author
  1. Columbia University, United States

Abstract

Recurrent neural networks (RNNs) enable the production and processing of time-dependent signals such as those involved in movement and working memory. Classic gradient-based algorithms for training RNNs have been available for decades, but are inconsistent with biological features of the brain, such as causality and locality. We derive an approximation to gradient-based learning that comports with these constraints by requiring synaptic weight updates to depend only on local information about pre- and postsynaptic activities, in addition to a random feedback projection of the RNN output error. In addition to providing mathematical arguments for the effectiveness of the new learning rule, we show through simulations that it can be used to train an RNN to perform a variety of tasks. Finally, to overcome the difficulty of training over very large numbers of timesteps, we propose an augmented circuit architecture that allows the RNN to concatenate short-duration patterns into longer sequences.

Data availability

Code implementing the RFLO learning algorithm for the example shown in Figure 2 has been included as a source code file accompanying this manuscript.

Article and author information

Author details

  1. James M Murray

    Zuckerman Mind, Brain, and Behavior Institute, Columbia University, New York, United States
    For correspondence
    jm4347@columbia.edu
    Competing interests
    The authors declare that no competing interests exist.
    ORCID icon "This ORCID iD identifies the author of this article:" 0000-0003-3706-4895

Funding

National Institutes of Health (DP5 OD019897)

  • James M Murray

National Science Foundation (DBI-1707398)

  • James M Murray

Gatsby Charitable Foundation

  • James M Murray

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

Copyright

© 2019, Murray

This article is distributed under the terms of the Creative Commons Attribution License permitting unrestricted use and redistribution provided that the original author and source are credited.

Metrics

  • 5,675
    views
  • 909
    downloads
  • 50
    citations

Views, downloads and citations are aggregated across all versions of this paper published by eLife.

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. James M Murray
(2019)
Local online learning in recurrent networks with random feedback
eLife 8:e43299.
https://doi.org/10.7554/eLife.43299

Share this article

https://doi.org/10.7554/eLife.43299

Further reading

    1. Medicine
    2. Neuroscience
    LeYuan Gu, WeiHui Shao ... HongHai Zhang
    Research Article

    The advent of midazolam holds profound implications for modern clinical practice. The hypnotic and sedative effects of midazolam afford it broad clinical applicability. However, the specific mechanisms underlying the modulation of altered consciousness by midazolam remain elusive. Herein, using pharmacology, optogenetics, chemogenetics, fiber photometry, and gene knockdown, this in vivo research revealed the role of locus coeruleus (LC)-ventrolateral preoptic nucleus noradrenergic neural circuit in regulating midazolam-induced altered consciousness. This effect was mediated by α1 adrenergic receptors. Moreover, gamma-aminobutyric acid receptor type A (GABAA-R) represents a mechanistically crucial binding site in the LC for midazolam. These findings will provide novel insights into the neural circuit mechanisms underlying the recovery of consciousness after midazolam administration and will help guide the timing of clinical dosing and propose effective intervention targets for timely recovery from midazolam-induced loss of consciousness.

    1. Neuroscience
    Ana Maria Ichim, Harald Barzan ... Raul Cristian Muresan
    Review Article

    Gamma oscillations in brain activity (30–150 Hz) have been studied for over 80 years. Although in the past three decades significant progress has been made to try to understand their functional role, a definitive answer regarding their causal implication in perception, cognition, and behavior still lies ahead of us. Here, we first review the basic neural mechanisms that give rise to gamma oscillations and then focus on two main pillars of exploration. The first pillar examines the major theories regarding their functional role in information processing in the brain, also highlighting critical viewpoints. The second pillar reviews a novel research direction that proposes a therapeutic role for gamma oscillations, namely the gamma entrainment using sensory stimulation (GENUS). We extensively discuss both the positive findings and the issues regarding reproducibility of GENUS. Going beyond the functional and therapeutic role of gamma, we propose a third pillar of exploration, where gamma, generated endogenously by cortical circuits, is essential for maintenance of healthy circuit function. We propose that four classes of interneurons, namely those expressing parvalbumin (PV), vasointestinal peptide (VIP), somatostatin (SST), and nitric oxide synthase (NOS) take advantage of endogenous gamma to perform active vasomotor control that maintains homeostasis in the neuronal tissue. According to this hypothesis, which we call GAMER (GAmma MEdiated ciRcuit maintenance), gamma oscillations act as a ‘servicing’ rhythm that enables efficient translation of neural activity into vascular responses that are essential for optimal neurometabolic processes. GAMER is an extension of GENUS, where endogenous rather than entrained gamma plays a fundamental role. Finally, we propose several critical experiments to test the GAMER hypothesis.