Shallow neural networks trained to detect collisions recover features of visual loom-selective neurons

  1. Baohua Zhou
  2. Zifan Li
  3. Sunnie Kim
  4. John Lafferty  Is a corresponding author
  5. Damon A Clark  Is a corresponding author
  1. Yale University, United States

Abstract

Animals have evolved sophisticated visual circuits to solve a vital inference problem: detecting whether or not a visual signal corresponds to an object on a collision course. Such events are detected by specific circuits sensitive to visual looming, or objects increasing in size. Various computational models have been developed for these circuits, but how the collision-detection inference problem itself shapes the computational structures of these circuits remains unknown. Here, inspired by the distinctive structures of LPLC2 neurons in the visual system of Drosophila, we build anatomically-constrained shallow neural network models and train them to identify visual signals that correspond to impending collisions. Surprisingly, the optimization arrives at two distinct, opposing solutions, only one of which matches the actual dendritic weighting of LPLC2 neurons. Both solutions can solve the inference problem with high accuracy when the population size is large enough. The LPLC2-like solutions reproduces experimentally observed LPLC2 neuron responses for many stimuli, and reproduces canonical tuning of loom sensitive neurons, even though the models are never trained on neural data. Thus, LPLC2 neuron properties and tuning are predicted by optimizing an anatomically-constrained neural network to detect impending collisions. More generally, these results illustrate how optimizing inference tasks that are important for an animal's perceptual goals can reveal and explain computational properties of specific sensory neurons.

Data availability

Code to perform all simulations in this paper and to reproduce all figures is available at http://www.github.com/ClarkLabCode/LoomDetectionANN.

Article and author information

Author details

  1. Baohua Zhou

    Department of Molecular, Cellular and Developmental Biology, Yale University, New Haven, United States
    Competing interests
    The authors declare that no competing interests exist.
  2. Zifan Li

    Department of Statistics and Data Science, Yale University, New Haven, United States
    Competing interests
    The authors declare that no competing interests exist.
  3. Sunnie Kim

    Department of Statistics and Data Science, Yale University, New Haven, United States
    Competing interests
    The authors declare that no competing interests exist.
    ORCID icon "This ORCID iD identifies the author of this article:" 0000-0002-8901-7233
  4. John Lafferty

    Department of Statistics and Data Science, Yale University, New Haven, United States
    For correspondence
    john.lafferty@yale.edu
    Competing interests
    The authors declare that no competing interests exist.
  5. Damon A Clark

    Department of Molecular, Cellular and Developmental Biology, Yale University, New Haven, United States
    For correspondence
    damon.clark@yale.edu
    Competing interests
    The authors declare that no competing interests exist.
    ORCID icon "This ORCID iD identifies the author of this article:" 0000-0001-8487-700X

Funding

National Institutes of Health (R01EY026555)

  • Baohua Zhou
  • Damon A Clark

National Science Foundation (CCF-1839308)

  • Baohua Zhou
  • John Lafferty
  • Damon A Clark

National Science Foundation (DMS-1513594)

  • John Lafferty

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

Copyright

© 2022, Zhou et al.

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

  • 1,516
    views
  • 206
    downloads
  • 23
    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. Baohua Zhou
  2. Zifan Li
  3. Sunnie Kim
  4. John Lafferty
  5. Damon A Clark
(2022)
Shallow neural networks trained to detect collisions recover features of visual loom-selective neurons
eLife 11:e72067.
https://doi.org/10.7554/eLife.72067

Share this article

https://doi.org/10.7554/eLife.72067

Further reading

    1. Neuroscience
    Franziska Auer, Katherine Nardone ... David Schoppik
    Research Article

    Cerebellar dysfunction leads to postural instability. Recent work in freely moving rodents has transformed investigations of cerebellar contributions to posture. However, the combined complexity of terrestrial locomotion and the rodent cerebellum motivate new approaches to perturb cerebellar function in simpler vertebrates. Here, we adapted a validated chemogenetic tool (TRPV1/capsaicin) to describe the role of Purkinje cells — the output neurons of the cerebellar cortex — as larval zebrafish swam freely in depth. We achieved both bidirectional control (activation and ablation) of Purkinje cells while performing quantitative high-throughput assessment of posture and locomotion. Activation modified postural control in the pitch (nose-up/nose-down) axis. Similarly, ablations disrupted pitch-axis posture and fin-body coordination responsible for climbs. Postural disruption was more widespread in older larvae, offering a window into emergent roles for the developing cerebellum in the control of posture. Finally, we found that activity in Purkinje cells could individually and collectively encode tilt direction, a key feature of postural control neurons. Our findings delineate an expected role for the cerebellum in postural control and vestibular sensation in larval zebrafish, establishing the validity of TRPV1/capsaicin-mediated perturbations in a simple, genetically tractable vertebrate. Moreover, by comparing the contributions of Purkinje cell ablations to posture in time, we uncover signatures of emerging cerebellar control of posture across early development. This work takes a major step towards understanding an ancestral role of the cerebellum in regulating postural maturation.

    1. Neuroscience
    Gáspár Oláh, Rajmund Lákovics ... Gábor Tamás
    Research Article

    Human-specific cognitive abilities depend on information processing in the cerebral cortex, where the neurons are significantly larger and their processes longer and sparser compared to rodents. We found that, in synaptically connected layer 2/3 pyramidal cells (L2/3 PCs), the delay in signal propagation from soma to soma is similar in humans and rodents. To compensate for the longer processes of neurons, membrane potential changes in human axons and/or dendrites must propagate faster. Axonal and dendritic recordings show that the propagation speed of action potentials (APs) is similar in human and rat axons, but the forward propagation of excitatory postsynaptic potentials (EPSPs) and the backward propagation of APs are 26 and 47% faster in human dendrites, respectively. Experimentally-based detailed biophysical models have shown that the key factor responsible for the accelerated EPSP propagation in human cortical dendrites is the large conductance load imposed at the soma by the large basal dendritic tree. Additionally, larger dendritic diameters and differences in cable and ion channel properties in humans contribute to enhanced signal propagation. Our integrative experimental and modeling study provides new insights into the scaling rules that help maintain information processing speed albeit the large and sparse neurons in the human cortex.