神经科学与认知科学 突破级 暂无讲解视频
发表时间
2025-12-26
arXiv
2512.21881

收录解读

问题与背景 Foundation modeling for fMRI faces a hard tradeoff: atlas-based methods are efficient but lose fine spatial detail, while voxel-level methods preserve fidelity at prohibitive memory and training cost.

方法/新意 SLIM-Brain proposes a more data- and training-efficient fMRI foundation-modeling approach that explicitly targets this bottleneck, aiming to retain useful spatial structure without the full cost of naive voxel-level scaling.

意义/放在仓库中的位置 This fits squarely within the repository's NeuroAI mainline. It is valuable because it tackles a structural bottleneck in fMRI foundation modeling rather than offering just another narrow downstream benchmark.

局限/为何不更高 It is still a preprint, and its longer-term importance depends on whether the efficiency/generalization tradeoff remains favorable as stronger large-scale baselines arrive.

链接