topo-omni-brain-topographic-multimodal

name	topo-omni-brain-topographic-multimodal
description	Deep topographic multimodal model (Topo-Omni) for discovering functionally selective brain regions with contiguous spatial organization across visual, auditory, and language/cognitive modalities.
version	1.0.0
author	arXiv Paper Authors
arxiv_id	2606.09770v1
published_date	2026-06-08T00:00:00.000Z
activation_keywords	["topographic model","brain topography","multimodal brain regions","cortical organization","functional selectivity","brain multimodal integration","spatial smoothness","Topo-Omni"]
categories	["neuroscience","computational neuroscience","deep learning","multimodal models","brain imaging"]
source	arXiv:2606.09770v1
paper_url	https://arxiv.org/abs/2606.09770
pdf_url	https://arxiv.org/pdf/2606.09770v1

Topo-Omni: Deep Topographic Multimodal Model for Brain Regions

Overview

Topo-Omni is a topographic multimodal model that enables visual, auditory, and language/cognitive processing to share a single contiguous in-silico sheet, addressing limitations of previous unimodal topographic models that produced fragmented maps.

Key Innovation

Single Contiguous Sheet Architecture: Unlike previous topographic models that spatially constrain each layer separately, Topo-Omni uses a unified spatial representation across all modalities, capturing both:

• Contiguity of cortical processing streams
• Integration across modalities

Core Methodology

Architecture Components

1. Foundation Model Initialization: Built by fine-tuning a pretrained foundation model
2. Spatial Smoothness Objective: Enforces topographic organization during training
3. Multimodal Integration: Visual, auditory, and language/cognitive processing share spatial coordinates

Training Approach

• Spatial Smoothness Constraint: Nearby neurons in the model should share similar response profiles
• Contiguous Processing: Ensures smooth transitions between modalities in the spatial layout
• Cross-modal Consistency: Clusters align with human neuroimaging findings

Key Findings

Biological Plausibility

• Develops clusters consistent with human neuroimaging from sensory to cognitive systems
• Reproduces systematic spatial organization observed in cortex
• Captures functional selectivity patterns matching known brain regions

Novel Contributions

1. Unified Spatial Map: First model to use single contiguous sheet for all modalities
2. Cross-modal Topography: Demonstrates integration across sensory and cognitive systems
3. Neuroimaging Alignment: Clusters correlate with actual human brain organization

Applications

Brain Research

• Discovering functionally selective brain regions computationally
• Understanding cortical organization principles
• Mapping multimodal integration zones

Clinical Applications

• Predicting functional deficits from lesion locations
• Understanding brain organization disorders
• Neuroimaging analysis tools