We introduce KumoRFM-2, the next iteration of a pre-trained foundation model for relational data. KumoRFM-2 supports in-context learning as well as fine-tuning and is applicable to a wide range of predictive tasks. In contrast to tabular foundation models, KumoRFM-2 natively operates on relational data, processing one or more connected tables simultaneously without manual table flattening or target variable generation, all while preserving temporal consistency. KumoRFM-2 leverages a large corpus of synthetic and real-world data to pre-train across four axes: the row and column dimensions at the individual table level, and the foreign key and cross-sample dimensions at the database level. In contrast to its predecessor, KumoRFM-2 injects task information as early as possible, enabling sharper selection of task-relevant columns and improved robustness to noisy data. Through extensive experiments on 41 challenging benchmarks and analysis around expressivity and sensitivity, we demonstrate that KumoRFM-2 outperforms supervised and foundational approaches by up to 8%, while maintaining strong performance under extreme settings of cold start and noisy data. To our knowledge, this is the first time a few-shot foundation model has been shown to surpass supervised approaches on common benchmark tasks, with performance further improving upon fine-tuning. Finally, while KumoRFM-1 was limited to small-scale in-memory datasets, KumoRFM-2 scales to billion-scale relational datasets.
Relational Learning in Pre-Trained Models: A Theory from Hypergraph Recovery Perspective
School of Intelligence Science and Technology, Peking University(北京大学智能科学与技术学院);Institute for Artificial Intelligence, Peking University(北京大学人工智能研究院);Pazhou Laboratory (Huangpu)(琶洲实验室(黄埔));University of Illinois Chicago(芝加哥大学伊利诺伊分校)
HeTa: Relation-wise Heterogeneous Graph Foundation Attack Model
School of Cyberspace, Hangzhou Dianzi University(电子科技大学信息学院);Data Security Governance Zhejiang Engineering Research Center, Hangzhou Dianzi University(浙江省数据安全治理工程研究中心);Beihang University(北航)
针对异质图神经网络攻击难以跨模型与跨图泛化的问题,提出 HeTa 基础攻击模型,利用共享关系感知攻击单元和逐关系序列化攻击实现可迁移、易适配的强攻击。
Position: Topological Deep Learning is the New Frontier for Relational Learning
The University of Manchester(曼彻斯特大学);Imperial College London(伦敦帝国学院);University of Oxford(牛津大学);Stanford University(斯坦福大学);BlueLightAI Inc(蓝光人工智能公司);University at Albany(奥尔巴尼大学);Tsinghua University(清华大学);University of San Francisco(旧金山大学);University of Salzburg(萨尔茨堡大学);University of Cambridge(剑桥大学);Sapienza University of Rome(罗马第一大学);University of Tennessee(田纳西大学);UC Santa Barbara(加利福尼亚大学圣巴巴拉分校);University of Hawaiʻi at Mānoa(夏威夷大学马诺阿分校);IBM Corporation(国际商业机器公司(IBM));Helmholtz Munich(慕尼黑亥姆霍兹中心);Technical University of Munich(慕尼黑工业大学)
Relational In-Context Learning via Synthetic Pre-training with Structural Prior
Institute for Artificial Intelligence, Peking University(北京大学人工智能研究院);University of Illinois at Urbana-Champaign(伊利诺伊大学香槟分校);Institute of Computing Technology, Beijing University of Post(北京邮电大学计算机学院);State Key Laboratory of General Artificial Intelligence(通用人工智能国家重点实验室)
SPR: Toward a Graph Foundation Model for Transferable Graph Cognition via Spectral Patterns and Relational Geometry
School of Computer Science and Technology(计算机科学与技术学院);Tianjin University(天津大学);School of Artificial Intelligence(人工智能学院);Hebei University of Technology(河北工业大学);Department of Health Technology and Informatics(健康技术与信息学院);Department of Data Science and Artificial Intelligence(数据科学与人工智能系);The Hong Kong Polytechnic University(香港理工大学)
RelationMatch: Matching In-batch Relationships for Semi-supervised Learning
IIIS, Tsinghua University(清华大学智能产业研究院(注:IIIS为Institute for Interdisciplinary Information Sciences的缩写,常见译名为交叉信息研究院));Department of Math, Tsinghua University(清华大学数学系)