analiza aiqbench
This study introduces a modality-agnostic framework that maps Transformer attention topologies onto human intrinsic connectivity networks, revealing that organizational alignment does not strictly correlate with model accuracy or standard tuning procedures. The analysis of 151 models uncovers counterintuitive scaling inversions and specific architectural behaviors, such as DINOv2's reduced alignment, offering a novel graph-based metric for cross-domain comparison. For researchers, this work shifts the focus from neural mechanism inference to structural topology, providing a task-free lens to evaluate how different model objectives map to human brain networks.
A Unified Geometric Space for Topological Alignment Between Transformer-Based Models and Human Brain Networks
arXiv:2510.24342v2 Announce Type: replace Abstract: Prior brain-AI alignment studies are typically constrained by specific inputs and tasks, limiting their ability to capture organizational properties across models with different modalities. In this work, we focus on Transformer-based models and introduce a brain-model topological alignment space. Rather than inferring alignment from neural mechanisms, we examine it through graph-based organizational properties, mapping the intrinsic spatial attention topology of a model onto canonical human intrinsic connectivity networks (ICNs). This enables a modality-agnostic and task-free comparison across vision, language, and multimodal systems at the level of organizational properties. Analyzing 151 Transformer-based models across these modalities and scales, we observe a continuous arc-shaped distribution, reflecting varying degrees of topological alignment. Consistent with their training objectives, models optimized for global semantic abstraction were associated more closely with higher-order ICNs, while local detail-focused models associated with low-level ICNs. More surprisingly, we uncovered non-intuitive phenomena: DINOv2 exhibited reduced alignment compared to its predecessors, distilled DeiT models showed a counterintuitive scaling inversion where larger models aligned less well with higher-order ICNs, and fine-tuning as well as instruction tuning had limited effect on alignment. Furthermore, topological alignment scores showed non-significant correlation with ImageNet-1K Top-1 accuracy in 30 vision Transformers (r=0.266, p=0.156). This work provides a new quantitative perspective for comparing the organizational properties of Transformer-based models through brain-referenced topological mapping.
Abstract:Prior brain-AI alignment studies are typically constrained by specific inputs and tasks, limiting their ability to capture organizational properties across models with different modalities. In this work, we focus on Transformer-based models and introduce a brain-model topological alignment space. Rather than inferring alignment from neural mechanisms, we examine it through graph-based organizational properties, mapping the intrinsic spatial attention topology of a model onto canonical human intrinsic connectivity networks (ICNs). This enables a modality-agnostic and task-free comparison across vision, language, and multimodal systems at the level of organizational properties. Analyzing 151 Transformer-based models across these modalities and scales, we observe a continuous arc-shaped distribution, reflecting varying degrees of topological alignment. Consistent with their training objectives, models optimized for global semantic abstraction were associated more closely with higher-order ICNs, while local detail-focused models associated with low-level ICNs. More surprisingly, we uncovered non-intuitive phenomena: DINOv2 exhibited reduced alignment compared to its predecessors, distilled DeiT models showed a counterintuitive scaling inversion where larger models aligned less well with higher-order ICNs, and fine-tuning as well as instruction tuning had limited effect on alignment. Furthermore, topological alignment scores showed non-significant correlation with ImageNet-1K Top-1 accuracy in 30 vision Transformers (r=0.266, p=0.156). This work provides a new quantitative perspective for comparing the organizational properties of Transformer-based models through brain-referenced topological mapping.
Artificial Intelligence (cs.AI)
arXiv:2510.24342 [cs.AI]
(or arXiv:2510.24342v2 [cs.AI] for this version)
Text integral sub licența deschisă a publicației · 225 cuvinte
