跳转至

文章背景与核心概要

现有的神经基础模型通常在特定的侵入式记录模态（如颅内脑电图 iEEG 或单神经元脉冲活动）上进行孤立训练，忽视了跨信号联合预训练所带来的优势。本文介绍了 iBrain，这是一个新颖的统一基础模型，旨在通过联合学习各种侵入式神经记录来弥补这一空白。

该模型通过利用特定信号编码器以及共享的时空 Transformer 主干网络，并在超过 7,000 小时的异构数据上使用掩码信号重建和通道视角对齐进行预训练，在多个基准测试中实现了 SOTA（最先进）的性能，同时表现出卓越的跨场景迁移能力和数据效率。

iBrain: A Unified Foundation Model Reading the Brain from Surface to Spikes

arXiv ID: arXiv:2609.06960 [cs.AI]
Submitted: September 7, 2026
Authors: Ying Chen, Tiou Wang, Zhifeng Yue

📌 Summary

Existing neural foundation models are typically trained in silos on specific invasive recording modalities, such as intracranial EEG (iEEG) or single-neuron spiking activity, ignoring the benefits of cross-signal joint pretraining. iBrain is a novel unified foundation model designed to bridge this gap by jointly learning from diverse invasive neural recordings.

现有的神经基础模型通常在特定的侵入式记录模态（如颅内脑电图 iEEG 或单神经元脉冲活动）上进行孤立训练，忽视了跨信号联合预训练所带来的优势。iBrain 是一个新颖的统一基础模型，旨在通过联合学习各种侵入式神经记录来弥补这一空白。

By leveraging signal-specific encoders alongside a shared spatiotemporal Transformer backbone, and pretraining on over 7,000 hours of heterogeneous data using masked signal reconstruction and channel-view alignment, iBrain achieves state-of-the-art performance across multiple benchmarks while exhibiting exceptional cross-setting transferability and data efficiency.

通过利用特定信号编码器以及共享的时空 Transformer 主干网络，并在超过 7,000 小时的异构数据上使用掩码信号重建和通道视角对齐进行预训练，iBrain 在多个基准测试中实现了 SOTA 性能，同时表现出卓越的跨场景迁移能力和数据效率。

🔬 Abstract

Invasive neural recordings provide high-fidelity measurements of brain activity, with signals such as intracranial EEG (iEEG) and intracortical spiking activity capturing neural dynamics at different spatial and temporal scales. Yet existing neural foundation models have largely been developed independently for different invasive recording paradigms, leaving joint pretraining across heterogeneous invasive signals underexplored.

In this work, we introduce iBrain, a unified foundation model that jointly learns from iEEG and spiking activity. iBrain employs signal-specific encoders to accommodate their distinct signal characteristics and a shared spatiotemporal Transformer backbone to model dependencies across recording channels and time. We pretrain iBrain on over 7,000 hours of heterogeneous neural recordings using masked signal reconstruction and channel-view alignment, promoting contextual modeling of neural dynamics and robustness across different channels. iBrain consistently outperforms single-signal pretraining baselines and achieves state-of-the-art performance on multiple benchmarks. Further experiments demonstrate that iBrain exhibits transferability and data efficiency across diverse recording settings. These results highlight the potential of joint pretraining on heterogeneous invasive neural recordings to support scalable neural modeling and transferable representations across recording settings and downstream tasks.

侵入式神经记录提供了大脑活动的高保真测量，诸如颅内脑电图 (iEEG) 和皮层内脉冲活动等信号在不同的空间和时间尺度上捕捉神经动力学。然而，现有的神经基础模型大都针对不同的侵入式记录范式独立开发，跨异构侵入式信号的联合预训练在很大程度上仍未得到充分探索。

在这项工作中，我们推出了 iBrain，这是一个能够同时从 iEEG 和脉冲活动中进行联合学习的统一基础模型。iBrain 采用特定信号的编码器来适应其独特的信号特征，并使用共享的时空 Transformer 主干网络来建模跨记录通道和时间的依赖关系。我们使用掩码信号重建和通道视角对齐，在超过 7,000 小时的异构神经记录上对 iBrain 进行了预训练，从而促进了神经动力学的上下文建模以及跨不同通道的鲁棒性。iBrain 在各项基准测试中始终优于单信号预训练基线，并实现了 SOTA 性能。进一步的实验表明，iBrain 在各种记录设置中表现出极佳的迁移能力和数据效率。这些结果凸显了对异构侵入式神经记录进行联合预训练的潜力，能够支持可扩展的神经建模以及跨记录设置和下游任务的可迁移表征。

⚙️ Core Architecture & Methodology

  • Heterogeneous Signal Processing: Utilizes signal-specific encoders tailored to the distinct characteristics of surface-level iEEG and high-resolution intracortical spikes.
  • Spatiotemporal Transformer Backbone: A unified architecture designed to model complex dependencies across both spatial recording channels and temporal sequences.
  • Pretraining Strategy: Trained on >7,000 hours of data utilizing:
  • Masked Signal Reconstruction for robust contextual modeling.
  • Channel-View Alignment to align representations across disparate recording configurations.
  • 异构信号处理： 采用针对表层 iEEG 和高分辨率皮层内脉冲不同特征定制的信号专用编码器。
  • 时空 Transformer 主干网络： 一种统一的架构，旨在对空间记录通道和时间序列之间的复杂依赖关系进行建模。
  • 预训练策略： 在 >7,000 小时 的数据上进行训练，利用：
  • 掩码信号重建，用于稳健的上下文建模。
  • 通道视角对齐，以对齐不同记录配置之间的表征。