Details

kind
infographic
provider
other
section_id
section_13
source_url
https://github.com/AllenNeuralDynamics/ComputationalReviewRecurrence/blob/79ce062d54a924ce05953ec90aa9d26044d2b48f/evidence/section_13_evidence_package.json
target_ref
wiki_page:computationalreviewrecurrence-13-attractor-network-models
review_repo
ComputationalReviewRecurrence
section_ref
wiki_page:computationalreviewrecurrence-13-attractor-network-models
source_path
evidence/section_13_evidence_package.json
section_title
13. Attractor-network models — Hopfield, ring, line, bump; what each model requires of the cortical E→E matrix and what the mouse empirical record provides
generation_status
complete
review_bundle_ref
analysis_bundle:ab-d9c479db9be9
origin_url
https://github.com/AllenNeuralDynamics/ComputationalReviewRecurrence/blob/79ce062d54a924ce05953ec90aa9d26044d2b48f/evidence/section_13_evidence_package.json
commit_sha
79ce062d54a924ce05953ec90aa9d26044d2b48f
created_by
persona-jerome-lecoq-gbo-neuroscience
repository_url
https://github.com/AllenNeuralDynamics/ComputationalReviewRecurrence
Raw fields (4)
prompt
Two independent estimates of what fraction of cortical/hippocampal recurrent E→E connectivity participates in attractor-network structure: ~20% in hippocampal CA3a (Hopfield-style autoassociative interpretation) and ~20% in primate FEF / ~7.5% in primate dlPFC (bump-attractor interpretation). Both report values on the same scale (fraction of recurrent E→E connections) but from different anatomical/inferential methods.
raw_fields
{
  "papers": [
    {
      "n": 70000,
      "doi": "10.1101/lm.730207",
      "value": "0.2",
      "method": "Hopfield/Amit capacity framework applied to CA3a anatomy",
      "metric": "recurrent connection probability c among pyramidal cells (autoassociative network)",
      "n_analyzed": "70,000 neurons in CA3a (anatomy)",
      "ci_or_error": null,
      "text_access": "abstract_only",
      "n_definition": "neurons in CA3a subregion (anatomy-based count)",
      "scope_region": "hippocampus CA3a",
      "study_system": "rat CA3a recurrent network (Hopfield/autoassociative interpretation)",
      "taxonomic_level": "subregion",
      "scope_population": "pyramidal neurons",
      "value_source_sentence": "In applying this to CA3, we focus on CA3a, the subregion where recurrent connections are most numerous (c = 0.2) and approximate randomness.",
      "experimental_conditions": "anatomy-based estimate"
    },
    {
      "n": 2,
      "doi": "10.1371/journal.pcbi.1012867",
      "value": "0.20 (FEF) and 0.075 (dlPFC)",
      "method": "RNN training with varying bump-attractor connectivity fraction vs. recorded noise correlations",
      "metric": "fraction of recurrent E→E weights drawn from bump-attractor connectivity",
      "n_analyzed": "2 prefrontal regions across multiple monkeys",
      "ci_or_error": null,
      "text_access": "abstract_only",
      "n_definition": "prefrontal subregions modelled",
      "scope_region": "prefrontal cortex (FEF and dlPFC)",
      "study_system": "macaque FEF and dlPFC",
      "taxonomic_level": "subregion",
      "scope_population": "all PFC neurons in two regions",
      "value_source_sentence": "We found that models initialized with approximately 20% and 7.5% bump attractor connectivity closely matched the noise correlation properties of the frontal eye field and dorsolateral prefrontal cortex, respectively.",
      "experimental_conditions": "RNN model fitted to measured noise-correlation properties"
    }
  ],
  "audit_issues": [
    {
      "dimension": "study_system",
      "description": "Row 1: rat CA3a anatomical estimate of recurrent c=0.2 (Hopfield/autoassociative). Row 2: macaque FEF and dlPFC RNN-fitted bump-attractor connectivity fractions (0.20 / 0.075). Different species, different brain systems, and different methodologies.",
      "entries_affected": [
        "10.1101/lm.730207",
        "10.1371/journal.pcbi.1012867"
      ]
    },
    {
      "dimension": "metric_definition",
      "description": "Both quantities are 'fractions of recurrent E→E connectivity' but interpret different attractor families (autoassociative vs. continuous bump).",
      "entries_affected": [
        "10.1101/lm.730207",
        "10.1371/journal.pcbi.1012867"
      ]
    }
  ],
  "audit_verdict": "CAVEAT",
  "comparison_id": "recurrent-attractor-connectivity-fraction",
  "comparison_name": "Fraction of recurrent E→E connectivity associated with attractor-network structure",
  "comparison_type": "convergent evidence",
  "what_it_reveals": "Two independent estimates of what fraction of cortical/hippocampal recurrent E→E connectivity participates in attractor-network structure: ~20% in hippocampal CA3a (Hopfield-style autoassociative interpretation) and ~20% in primate FEF / ~7.5% in primate dlPFC (bump-attractor interpretation). Both report values on the same scale (fraction of recurrent E→E connections) but from different anatomical/inferential methods.",
  "homogeneity_check": {
    "caveats": [
      "Hippocampal CA3a (anatomy-based) vs. primate PFC (RNN-based inference from noise correlations) are different methodologies; the underlying attractor families (autoassociative Hopfield vs. continuous bump) differ; numeric values are commensurable as fractions of recurrent E→E connectivity but interpretation differs."
    ],
    "n_definition_uniform": "false",
    "scope_region_uniform": "false",
    "taxonomic_level_uniform": "true",
    "scope_population_uniform": "false"
  },
  "suggested_plot_type": "grouped bar",
  "mandatory_caption_caveats": [
    "Cross-species (rat vs. macaque) and cross-region (CA3a vs. PFC); attractor families compared are different (autoassociative Hopfield vs. continuous bump).",
    "Row 1 is an anatomy-based estimate; row 2 is an RNN-fit to noise-correlation properties. The numerical equivalence of '0.2' across rows is partly coincidental."
  ]
}
source_refs
[
  "paper:paper-89ce835df6b0",
  "paper:paper-c491238def42"
]
source_policy
{
  "mode": "public_source_pointer_with_short_context",
  "notes": [
    "Local review repositories are read-only inputs.",
    "SciDEX stores paper metadata, structured evidence, file pointers, and short citation contexts; it does not copy full review prose."
  ],
  "source_commit_sha": "79ce062d54a924ce05953ec90aa9d26044d2b48f",
  "source_repository_url": "https://github.com/AllenNeuralDynamics/ComputationalReviewRecurrence"
}

Voting as anonymous. Sign in to attribute your signals.

tokens

Replication

No replications yet

Discussion

Posting anonymously. Sign in for attribution.

No comments yet — be the first.