Recurrent Predictive Learning (RPL), a recurrent joint-embedding predictive arc…

Recurrent Predictive Learning (RPL), a recurrent joint-embedding predictive architecture, computes prediction errors only in latent representation space (not input space) and learns sequence representations that resemble successor-like codes reported in human V1 and abstract sequence codes reported in macaque PFC — a circuit-centric theory framework that can be mapped onto canonical cortical microcircuits.

{
  "n": 0,
  "doi": "10.1101/2025.11.25.690220",
  "claim": "Recurrent Predictive Learning (RPL), a recurrent joint-embedding predictive architecture, computes prediction errors only in latent representation space (not input space) and learns sequence representations that resemble successor-like codes reported in human V1 and abstract sequence codes reported in macaque PFC — a circuit-centric theory framework that can be mapped onto canonical cortical microcircuits.",
  "cite_key": "Mohammadi2025",
  "evidence": "JEPA-style theoretical model; representation comparisons with human V1 and macaque PFC; mapping to canonical microcircuit.",
  "effect_size": "model learns successor-like sequence representations matching human V1 and macaque PFC patterns",
  "text_access": "abstract_only",
  "study_system": "Theory (mapped to canonical cortical microcircuit)",
  "argument_role": "supporting",
  "replication_status": "theory",
  "claim_source_sentence": "Specifically, we introduce recurrent predictive learning (RPL), a recurrent joint-embedding predictive architecture inspired by self-supervised machine learning, that learns abstract representations of object identity and their dynamics and predicts future object motion. Crucially, the model learns sequence representations that resemble successor-like representations observed in the primary visual cortex of humans.",
  "source_provenance_status": "non_substring_match",
  "replication_evidence_dois": [],
  "effect_size_source_sentence": "the model learns sequence representations that resemble successor-like representations observed in the primary visual cortex of humans."
}

[
  "paper:paper-fbd1daca6501"
]

Specifically, we introduce recurrent predictive learning (RPL), a recurrent joint-embedding predictive architecture inspired by self-supervised machine learning, that learns abstract representations of object identity and their dynamics and predicts future object motion. Crucially, the model learns sequence representations that resemble successor-like representations observed in the primary visual cortex of humans.

[
  {
    "ref": "paper:paper-fbd1daca6501"
  }
]

{
  "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"
}