ReGraph: A Computational Account of Emergent Generalization in the "what" and "where" Dual Visual Streams
Abstract
Where generalization capacity--the ability to extract context-invariant relational structures--first emerges remains a central question in AI and neuroscience. The foundation for this capacity lies upstream of the hippocampus, within the entorhinal cortex, where parallel pathways dissociate relational structure in the medial entorhinal cortex (MEC) from sensory content in the lateral entorhinal cortex. However, as Eichenbaum argued, such factorization likely originates earlier, driven by the segregation of the dorsal ('where') and ventral ('what') visual streams. Supporting this, grid-like firing patterns--a signature of MEC (context-invariant codes)--also appear in preceding neocortical regions along the dorsal pathway. Yet, how such representations are computationally formed along upstream pathways remains unknown. To investigate this in silico, we developed ReGraph, a recurrent dual-stream graph model with biological inductive biases, including retina-driven stream-specialized encoding, dorsal-to-ventral modulation, and dynamic lateral connectivity. Trained on the action benchmark Something-Something V2, ReGraph revealed a pathway-specific emergence of relational mapping: context-invariant codes and grid-like spatial bases uniquely co-emerged along the extended dorsal stream. In contrast, their absence in single-stream, unmodulated variants, and standard baselines implies that these inductive biases are prerequisites for relational structures. Crucially, our post-hoc analyses demonstrated that these grid-like bases serve as reusable routing templates for information processing via lateral connectivity. Together, our findings provide a computational account that generalization may not be a faculty that emerges abruptly within a dedicated region, but a property that already takes shape as sensory information is parsed into factorized streams of hierarchical visual processing.
Related
- Can the Human Plastisphere Account for the Rise of Type 1 Diabetes: A Computational Molecular Mimicry Study
- A single computational objective can produce specialization of streams in visual cortex
- Emergent Generalization by Representation Learning in Artificial Neural Networks
- Overlapping perceptual benefits of temporal attention in aperiodic and periodic streams are mediated by delta phase alignment, but not alpha suppression, in sensory circuits
- How Institutional Distrust Breeds One-Sided Leadership: A Consistency-Based Account
BibTeX
@misc{kang2026regraph,
title = {{ReGraph: A Computational Account of Emergent Generalization in the "what" and "where" Dual Visual Streams}},
author = {Hyewon Kang and Jungmin Lee and Ilgyu Lee and Seok-Jun Hong},
howpublished = {arXiv},
year = {2026},
month = oct,
doi = {10.48550/arxiv.2610.07962},
eprint = {2610.07962},
archivePrefix = {arXiv},
url = {https://arxiv.org/abs/2610.07962},
}
RIS
TY - UNPB
TI - ReGraph: A Computational Account of Emergent Generalization in the "what" and "where" Dual Visual Streams
AU - Hyewon Kang
AU - Jungmin Lee
AU - Ilgyu Lee
AU - Seok-Jun Hong
PY - 2026
DA - 2026/10/06
PB - arXiv
DO - 10.48550/arxiv.2610.07962
UR - https://arxiv.org/abs/2610.07962
AB - Where generalization capacity--the ability to extract context-invariant relational structures--first emerges remains a central question in AI and neuroscience. The foundation for this capacity lies upstream of the hippocampus, within the entorhinal cortex, where parallel pathways dissociate relational structure in the medial entorhinal cortex (MEC) from sensory content in the lateral entorhinal cortex. However, as Eichenbaum argued, such factorization likely originates earlier, driven by the segregation of the dorsal ('where') and ventral ('what') visual streams. Supporting this, grid-like firing patterns--a signature of MEC (context-invariant codes)--also appear in preceding neocortical regions along the dorsal pathway. Yet, how such representations are computationally formed along upstream pathways remains unknown. To investigate this in silico, we developed ReGraph, a recurrent dual-stream graph model with biological inductive biases, including retina-driven stream-specialized encoding, dorsal-to-ventral modulation, and dynamic lateral connectivity. Trained on the action benchmark Something-Something V2, ReGraph revealed a pathway-specific emergence of relational mapping: context-invariant codes and grid-like spatial bases uniquely co-emerged along the extended dorsal stream. In contrast, their absence in single-stream, unmodulated variants, and standard baselines implies that these inductive biases are prerequisites for relational structures. Crucially, our post-hoc analyses demonstrated that these grid-like bases serve as reusable routing templates for information processing via lateral connectivity. Together, our findings provide a computational account that generalization may not be a faculty that emerges abruptly within a dedicated region, but a property that already takes shape as sensory information is parsed into factorized streams of hierarchical visual processing.
ER -