发表机构
Sorbonne Université; CNRS; ISIR; MLIA; Inria; I3S; Institut Universitaire de France (IUF); INSA-Lyon; Université Claude Bernard Lyon 1; Inserm; CREATIS(索邦大学; 法国国家科学研究中心; 巴黎索邦大学机器人与智能系统研究所; 机器学习与人工智能实验室; 法国国家信息与自动化研究所; 信息科学与系统实验室; 法国大学研究院; 里昂国立应用科学学院; 里昂第一大学; 法国国家健康与医学研究院; 图像、信号与智能系统研究中心)
机构由 AI 辅助整理,请以论文原文为准。AI 中文总结
针对中低数据下生理时间序列分类的泛化难题,提出适配时序数据的TSPFN基础模型,经14万真实生理序列预训练后,在多基准上优于TabPFN等模型,跨域泛化更优。
AI 中文摘要
在中低数据量场景下设计具备有效泛化能力的模型仍是医学机器学习的核心挑战,尤其针对生理时间序列分类任务。尽管表格基础模型如TabPFN通过上下文学习为传统微调提供了颇具吸引力的替代方案,但这类模型并非为捕捉生理信号固有时序依赖而设计。本文提出TSPFN,一款针对时间序列数据重新设计TabPFN架构的基础模型,其整合结构化时序表示与位置嵌入,以捕捉样本内时序及通道依赖。为充分发挥其时-空设计优势,该模型在跨多个医学领域的14万个真实世界生理时间序列上进行预训练,得到一个统一、可泛化的框架,能够学习医学时间序列的特性。在不同生理基准上开展的实验表明,TSPFN始终优于标准表格基线模型与TabPFN,且相比专用深度时间序列模型实现了更优的跨领域泛化。本文所有实验、消融研究及预处理方案均公开于指定网址。
英文摘要
Designing models that generalize effectively in low- to medium-data regimes remains a primary challenge in medical machine learning, particularly for physiological time-series classification. While tabular foundation models such as TabPFN offer an attractive alternative to conventional fine-tuning through in-context learning, they are not designed to capture the temporal dependencies inherent to physiological signals. ~In this paper, we introduce TSPFN, a foundation model that redesigns TabPFN's architecture for time series data. TSPFN integrates structured temporal representations and positional embeddings to capture intra-sample temporal and channel dependencies. To fully leverage its spatio-temporal design, the model is pretrained on 140,000 real-world physiological time series across multiple medical domains. This yields a unified, generalizable framework capable of learning the specificities of medical time series. Experiments across diverse physiological benchmarks demonstrate that TSPFN consistently outperforms standard tabular baselines and TabPFN, and achieves superior cross-domain generalization compared to specialized deep time-series models. All our experiments, ablation studies, and pre-processing scheme are publicly available at https://github.com/Jeremstym/TSPFN
CommentsAccepted at STACOM 2026 (MICCAI Workshop). 10 pages, 3 figures