CoChat

The DOI matches its Crossref record; 96 of the 96 references it cites were found in a registry or an index.

Predictive Feedback Can Account for Biphasic Responses in the Lateral Geniculate Nucleus

Open access
Jehee, Janneke F. M., Ballard, Dana H. · 2009 · PLoS Computational Biology

Abstract

Biphasic neural response properties, where the optimal stimulus for driving a neural response changes from one stimulus pattern to the opposite stimulus pattern over short periods of time, have been described in several visual areas, including lateral geniculate nucleus (LGN), primary visual cortex (V1), and middle temporal area (MT). We describe a hierarchical model of predictive coding and simulations that capture these temporal variations in neuronal response properties. We focus on the LGN-V1 circuit and find that after training on natural images the model exhibits the brain's LGN-V1 connectivity structure, in which the structure of V1 receptive fields is linked to the spatial alignment and properties of center-surround cells in the LGN. In addition, the spatio-temporal response profile of LGN model neurons is biphasic in structure, resembling the biphasic response structure of neurons in cat LGN. Moreover, the model displays a specific pattern of influence of feedback, where LGN receptive fields that are aligned over a simple cell receptive field zone of the same polarity decrease their responses while neurons of opposite polarity increase their responses with feedback. This phase-reversed pattern of influence was recently observed in neurophysiology. These results corroborate the idea that predictive feedback is a general coding strategy in the brain.

Overview

The claims and statistics of this paper haven’t been checked yet.

Free: it reads the open-access text and checks the first 5 claims that cite a source we can check. Create a free account to check every claim.

A few free checks a day · papers already verified open instantly · results are public

Integrity signals

Registry

Crossref · 10.1371/journal.pcbi.1000373

Cites

96 of 96 references verified0 citation errors · 0 not checkable

Statistics

Figures

Figure 1 Hierarchical model for predictive coding. Higher-level coding units attempt to predict the responses of units in the next lower level via feedback connections, while lower-level error detectors signal the difference between the prediction and the actual input. Feedforward connections encode the synaptic weights represented by a matrix . Higher-level units maintain the current estimate of the input signal and convey the top-down prediction to the lower level via feedback connections. Difference detectors compute the difference between current activity and the top-down prediction .
Figure 1 Hierarchical model for predictive coding. Higher-level coding units attempt to predict the responses of units in the next lower level via feedback connections, while lower-level error detectors signal the difference between the prediction and the actual input. Feedforward connections encode the synaptic weights represented by a matrix . Higher-level units maintain the current estimate of the input signal and convey the top-down prediction to the lower level via feedback connections. Difference detectors compute the difference between current activity and the top-down prediction .CC BY · PubMed Central
Figure 2 Receptive fields of model V1 units after training on natural images. (A) Examples of natural images used for training. The square denotes model V1 receptive field size. (B) V1 receptive fields after training. Plots are scaled in magnitude so that each fills the gray scale, but with zero always represented by the same gray level. Black depicts off-regions in the model V1 receptive field, white depicts on-regions.
Figure 2 Receptive fields of model V1 units after training on natural images. (A) Examples of natural images used for training. The square denotes model V1 receptive field size. (B) V1 receptive fields after training. Plots are scaled in magnitude so that each fills the gray scale, but with zero always represented by the same gray level. Black depicts off-regions in the model V1 receptive field, white depicts on-regions.CC BY · PubMed Central
Figure 3 Connection weights after training. The figure depicts learned connection weights from 64 LGN off-center type units and from 64 LGN on-center type units to one representative V1 unit. Red: connection weights from on-center type cells, blue: connection weights from off-center type cells. Brighter values indicate higher connection weights. The value zero is represented by the color black. The on- and off-center units are spatially aligned with the on- and off-zones of the model V1 receptive field.
Figure 3 Connection weights after training. The figure depicts learned connection weights from 64 LGN off-center type units and from 64 LGN on-center type units to one representative V1 unit. Red: connection weights from on-center type cells, blue: connection weights from off-center type cells. Brighter values indicate higher connection weights. The value zero is represented by the color black. The on- and off-center units are spatially aligned with the on- and off-zones of the model V1 receptive field.CC BY · PubMed Central
Figure 4 Spatio-temporal response of LGN on-center type cell. The response was mapped with the reverse correlation algorithm, using either non-biphasic retinal inputs (A–C,E) or biphasic retinal inputs (D). (A) Spatio-temporal response of an on-center type cell in model LGN. Responses were obtained by cross-correlating stimulus and response at the time intervals given below the figures. Red: response to bright stimulus at that location, blue: response to dark stimulus at that location. Note the change in sign after 50 milliseconds. Similar results were obtained for other LGN on-center type units. (B) Spatio-temporal response profile of on-center type cells in cat LGN obtained with the reverse correlation algorithm [1]. (C) The removal of feedback in the model causes the previously biphasic responses to disappear. (D) Temporal response profile of on-center type cell in a model with biphasic retinal inputs. Model activity is normalized by the initial response magnitudes. The biphasic response in LGN is more pronounced in the presence of predictive feedback compared to a situation in which the LGN response is fully determined by biphasic retinal input (for comparison, see e.g. [21]). (E) Average model LGN and V1 representations after reference stimuli consisting of bright stimulus regions have been presented. Black depicts off regions, white depicts on regions. When V1 predictions of the bright reference stimulus arrive in model LGN, they are compared against a new and unexpected stimulus representation. The difference between the predicted bright region and the second stimulus is negative, exciting LGN off-center type cells.
Figure 4 Spatio-temporal response of LGN on-center type cell. The response was mapped with the reverse correlation algorithm, using either non-biphasic retinal inputs (A–C,E) or biphasic retinal inputs (D). (A) Spatio-temporal response of an on-center type cell in model LGN. Responses were obtained by cross-correlating stimulus and response at the time intervals given below the figures. Red: response to bright stimulus at that location, blue: response to dark stimulus at that location. Note the change in sign after 50 milliseconds. Similar results were obtained for other LGN on-center type units. (B) Spatio-temporal response profile of on-center type cells in cat LGN obtained with the reverse correlation algorithm [1]. (C) The removal of feedback in the model causes the previously biphasic responses to disappear. (D) Temporal response profile of on-center type cell in a model with biphasic retinal inputs. Model activity is normalized by the initial response magnitudes. The biphasic response in LGN is more pronounced in the presence of predictive feedback compared to a situation in which the LGN response is fully determined by biphasic retinal input (for comparison, see e.g. [21]). (E) Average model LGN and V1 representations after reference stimuli consisting of bright stimulus regions have been presented. Black depicts off regions, white depicts on regions. When V1 predictions of the bright reference stimulus arrive in model LGN, they are compared against a new and unexpected stimulus representation. The difference between the predicted bright region and the second stimulus is negative, exciting LGN off-center type cells.CC BY · PubMed Central
Figure 5 Effects of feedback on LGN on-center and off-center type cells. Dashed: probability that the LGN cell coding for this location is active (i.e. response>0) in the first feedforward sweep of the model when a V1 region will subsequently be selected that codes for the same or opposite polarity, blank: probability that the LGN cell is active after the first feedforward-feedback pass (i.e. when feedback exerts its effect) when a V1 region is selected that codes for the same or opposite polarity. Red: on-center type cell, blue: off-center type cell. The results were obtained after presenting the model with a white-noise stimulus every 100 milliseconds for a total of 10,000 images. Comparing initial feedforward activity with subsequent LGN activity shows that feedback has a negative influence on cells of similar sign, and a positive influence on cells of opposite sign. Thus, the probability that an LGN off-cell is active increases after feedback from a V1 on-region (lower right, blue), and the probability that an LGN on-type cell is active decreases after feedback from an on-region in V1 (upper left, red).
Figure 5 Effects of feedback on LGN on-center and off-center type cells. Dashed: probability that the LGN cell coding for this location is active (i.e. response>0) in the first feedforward sweep of the model when a V1 region will subsequently be selected that codes for the same or opposite polarity, blank: probability that the LGN cell is active after the first feedforward-feedback pass (i.e. when feedback exerts its effect) when a V1 region is selected that codes for the same or opposite polarity. Red: on-center type cell, blue: off-center type cell. The results were obtained after presenting the model with a white-noise stimulus every 100 milliseconds for a total of 10,000 images. Comparing initial feedforward activity with subsequent LGN activity shows that feedback has a negative influence on cells of similar sign, and a positive influence on cells of opposite sign. Thus, the probability that an LGN off-cell is active increases after feedback from a V1 on-region (lower right, blue), and the probability that an LGN on-type cell is active decreases after feedback from an on-region in V1 (upper left, red).CC BY · PubMed Central
Figure 6 Predictive feedback using the matching pursuit algorithm. (A) Model receptive fields (RFs) are represented as basis vectors. When the input (blue vector) arrives in model V1, a basis vector that has high overlap with the input is selected (red vector ). The V1 basis vector weighted by its response is then subtracted from the input and the selection-subtraction process is repeated on the residual LGN representation (green vector). (B,C) Model V1 prediction and residual LGN representation over time. (B) The blue vector represents the actual input, its prediction is represented by the red dot. (C) Black depicts off-regions, white depicts on-regions. A prediction is obtained by summing the selected V1 basis vectors weighted by their response. LGN difference detectors represent the error between V1 prediction and actual input. (B,C) Subsequent feedforward-feedback cycles refine the higher-level prediction of the input. Without predictive feedback, the model would represent just the initial, less accurate prediction.
Figure 6 Predictive feedback using the matching pursuit algorithm. (A) Model receptive fields (RFs) are represented as basis vectors. When the input (blue vector) arrives in model V1, a basis vector that has high overlap with the input is selected (red vector ). The V1 basis vector weighted by its response is then subtracted from the input and the selection-subtraction process is repeated on the residual LGN representation (green vector). (B,C) Model V1 prediction and residual LGN representation over time. (B) The blue vector represents the actual input, its prediction is represented by the red dot. (C) Black depicts off-regions, white depicts on-regions. A prediction is obtained by summing the selected V1 basis vectors weighted by their response. LGN difference detectors represent the error between V1 prediction and actual input. (B,C) Subsequent feedforward-feedback cycles refine the higher-level prediction of the input. Without predictive feedback, the model would represent just the initial, less accurate prediction.CC BY · PubMed Central

Verification

Claims the paper makes

The claims and statistics of this paper haven’t been checked yet.

Free: it reads the open-access text and checks the first 5 claims that cite a source we can check. Create a free account to check every claim.

A few free checks a day · papers already verified open instantly · results are public

Claims other papers make about this one

Each is checked against this paper’s own text: does it say what the citing paper says it does?

Self-check

Percentages, sums and reported differences re-computed from the paper’s own numbers.

The self-check needs the paper’s full text.

Citations

Cites

Show all 95 references

Research works printed but not linked: 1

  1. DL Ringach, 1997, Dynamics of orientation tuning in macaque primary visual cortex., Nature, 387, 281, 10.1038/387281a0 · DOI · Paper page

Cited by

See all 75

Metadata

Authors
Jehee, J. F. M., & Ballard, D. H.
Affiliations
  • Vanderbilt University (US)
  • University of Rochester (US)
Corresponding author
  • Jehee, Janneke F. M.
  • Ballard, Dana H.
Venue
PLoS Computational Biology
Published
2009-05-01
Volume
5
Issue
5
Pages
e1000373
ISSN
  • 1553-734X (ISSN-L)
  • 1553-7358
OpenAlex
W2141564777
Cited by
75
References
95
Citations by year
2026: 1 · 2025: 3 · 2024: 4 · 2023: 2 · 2022: 2 · 2021: 6 · 2020: 3 · 2019: 4 · 2018: 10 · 2017: 3 · 2016: 13 · 2015: 2 · 2014: 3 · 2013: 7 · 2012: 6
FWCI
1.94
Type
journal article
Language
en
Open access
Licence
cc-by
Locations
  • PLoS Computational Biology (publishedVersion, cc-by)
  • PubMed Central (publishedVersion, cc-by)
  • PubMed (publishedVersion)
  • DOAJ (DOAJ: Directory of Open Access Journals) (submittedVersion, cc-by-sa)
  • Texas ScholarWorks (Texas Digital Library) (submittedVersion, cc-by)
Funders
  • Nederlandse Organisatie voor Wetenschappelijk Onderzoek
  • National Institutes of Health (EY05729)
  • National Eye Institute (R01EY005729, EY05729)
  • National Center for Research Resources (R01RR009283)
Topics
  • Life Sciences › Neuroscience › Cognitive Neuroscience › Visual perception and processing mechanisms
  • Life Sciences › Neuroscience › Cognitive Neuroscience › Neural dynamics and brain function
  • Physical Sciences › Computer Science › Computer Vision and Pattern Recognition › Visual Attention and Saliency Detection
Keywords
predictive coding, lateral geniculate nucleus, primary visual cortex, receptive fields, hierarchical models, phase reversal, visual cortex
MeSH terms
Algorithms, Animals, Artificial Intelligence, Brain, Cats, Computer Simulation, Feedback, Geniculate Bodies, Models, Neurological, Neurons, Visual Cortex

About these checks

Registry, retraction, Cites and statistics results are computed, not judged, with one exception: a printed reference that no index lists is read once by a model before it counts as a citation error. Claims are extracted by a model and kept only when their quote appears verbatim in the text; each cited source is then checked, with a model’s help, for whether it says what the claim relies on. Those verdicts can be wrong — read the claim’s quote and the cited source before relying on one.

Statistics and arithmetic are recomputed, not judged.

Confidence bands are shown only for fields calibrated on our annotated test set. None is calibrated yet, so no band is shown.

What counts as a claim ›How we measure confidence ›How we check this ›

Are you an author, or do you see a problem with this page? Request a review

How to cite

Cite CoChat for the checks on this page, and the paper for its own findings.

This page
CoChat. (2026). Citation check: Predictive Feedback Can Account for Biphasic Responses in the Lateral Geniculate Nucleus. https://app.cochat.ai/papers/10.1371/journal.pcbi.1000373
The paper
Jehee, J. F. M., & Ballard, D. H. (2009). Predictive Feedback Can Account for Biphasic Responses in the Lateral Geniculate Nucleus. PLoS Computational Biology. https://doi.org/10.1371/journal.pcbi.1000373

Data sources

Bibliographic data: Semantic Scholar, OpenAlex, Crossref.

Contains information from the Semantic Scholar Academic Graph Datasets, made available under the ODC Attribution License (ODC-BY 1.0).

Semantic Scholar’s data is described in: Kinney, R., et al. (2023). The Semantic Scholar Open Data Platform. arXiv:2301.10140.