neuralset-neuro-ai-framework

name	neuralset-neuro-ai-framework
description	NeuralSet unified Python framework for Neuro-AI research, harmonizing diverse neural recordings (fMRI, M/EEG, spikes) with deep learning embeddings
category	neuroscience
authors	["Jean-Rémi King","Corentin Bel","Linnea Evanson","Julien Gadonneix","Sophia Houhamdi","Jarod Lévy","Josephine Raugel","Andrea Santos Revilla","Mingfang Zhang","Julie Bonnaire","Charlotte Caucheteux","Alexandre Défossez","Théo Desbordes","Pablo Diego-Simón","Shubh Khanna","Juliette Millet","Pierre Orhan","Saarang Panchavati","Antoine Ratouchniak","Alexis Thual","Teon L. Brooks","Katelyn Begany","Yohann Benchetrit","Marlène Careil","Hubert Banville","Stéphane d'Ascoli","Simon Dahan","Jérémy Rapin"]
arxiv_id	2605.03169
submission_date	2026-05-04
doi	https://doi.org/10.48550/arXiv.2605.03169
github	https://github.com/neuralsign/neuralset
tags	["neuro-ai","python framework","data harmonization","fMRI","EEG","MEG","spike recordings","deep learning embeddings","lazy loading","memory-efficient","scalable infrastructure"]
activation_keywords	["neuro-ai","neural data preprocessing","fMRI EEG MEG harmonization","deep learning embeddings for neuroscience","lazy loading","memory-efficient neural data","PyTorch-ready neural data","neural dataset scaling","computational provenance"]

NeuralSet: A High-Performing Python Package for Neuro-AI

概述

NeuralSet是统一的Python框架，高效处理多样化神经记录（fMRI, M/EEG, spikes）和复杂实验刺激（文本、音频、视频），将神经预处理管道与预训练深度学习嵌入无缝集成。

核心创新

1. 模态无关的数据统一

问题：当前神经科学软件工具分散，按记录模态隔离，无法跨模态整合。

解决方案：

• 单一接口统一处理fMRI、EEG、MEG、spike recordings
• 标准化数据提取管道，消除模态差异
• 支持复杂实验刺激（文本、音频、视频）的嵌入生成

数据流架构：

原始数据 → 模态适配器 → 标准化提取 → 深度嵌入 → PyTorch张量
[fMRI, EEG, MEG, spikes] + [text, audio, video] → unified interface

2. 惰性加载与内存效率

机制：

• 解耦实验元数据与惰性数据提取
• 仅加载当前所需数据片段，避免全数据集内存占用
• 支持大规模自然数据集处理（GB级→TB级）

实现策略：

# 惰性加载伪代码
class NeuralDataset:
    def __init__(self, metadata_path):
        self.metadata = load_metadata(metadata_path)  # 小型元数据
        self.data_paths = parse_data_paths(metadata)
        self.lazy_loader = LazyLoader()  # 惰性加载器
    
    def get_sample(self, index):
        # 仅加载单个样本，不预加载全数据集
        return self.lazy_loader.load(self.data_paths[index])
    
    def __iter__(self):
        # 流式迭代，内存占用恒定
        for i in range(len(self)):
            yield self.get_sample(i)

内存优化：

• 元数据（KB级）与实际数据（GB级）分离