Topo-Omni AI model discovers new brain regions for landscapes and animals
AI uncovers previously unknown visual networks in the brain, validated with human data
Researchers led by Badr AlKhamissi developed Topo-Omni, a deep topographic multimodal model that unifies visual, auditory, and language/cognitive processing on a single contiguous in-silico sheet. Unlike previous topographic models that were unimodal and spatially constrained per layer, Topo-Omni is built by fine-tuning a pretrained foundation model with a spatial smoothness objective. This approach produces clusters across modalities that mirror the spatial organization seen in human cortical neuroimaging, from early sensory areas to higher cognitive regions. The model's architecture allows for selective driving or suppression of these clusters—reproducing effects akin to human intervention studies.
Beyond replicating known functional selectivity, Topo-Omni was used to screen for novel brain-like clusters in-silico. It discovered distinct networks responding to natural landscapes and animals—regions not previously cataloged in the model’s training. Critically, these newly identified networks were validated against real human fMRI data, confirming their existence in biological brains. This work demonstrates that a single spatial principle can organize representations across modalities and processing stages, offering a powerful framework for generating testable hypotheses about cortical organization. The paper is available as a preprint on arXiv.
- Topo-Omni fine-tunes a foundation model with a spatial smoothness objective to create a unified cortical sheet for vision, audition, and language
- The model's clusters match human neuroimaging data and can be selectively activated or suppressed to bias perception
- In-silico screening discovered novel landscape and animal networks that were subsequently validated in real human fMRI data
Why It Matters
A scalable AI model that predicts and discovers brain regions, accelerating neuroscience hypotheses and linking AI representations to cortical organization.