cs.AI

LimiX-2: A Contextual Mechanism Network Towards General Structured-Data Intelligence

Xingxuan Zhang, Gang Ren, Hao Yuan, Hao Zou, Hongze Tan, Hui Wang, Jianhao Song, Jiansheng Li, Jiayao Zhang, Jinghan Zhang, Kaifang Li, Lang Mo, Li Mao, Mingchao Hao, Nuo Xu, Rui Ding, Ruiji Zhang, Shuyang Li, Siyu Mei, Tianyang Zhang, Weiyang Mu, Yancheng Dong, Yongxian Wei, Yuan Xue, Yuanrui Wang, Yue He, Zijia Yang, Ziyun Li, Dongzhe Li, Fuqiang Wang, Jiandong Liu, Jiawei Chen, Jiaxin Du, Kaijie Cheng, Kehan Li, Lei Sun, Linjun Zhou, Ningbo Dai, Qi Wang, Renzhe Xu, Shaoxing Du, Shumeng Yang, Wang Lu, Wenjing Chu, Xiannan Huang, Xiaoyu Lin, Xing Ai, Xinyan Han, Xuanyue Li, Xuanyue Su, Xukun Zhang, Yan Lu, Yaxin Zhang, Yi Qin, Yifei Huang, Yihan Xu, Yongle Lv, Yuanyuan Jiang, Yushan Han, Peng Cui

arXiv: 2609.17488Published: September 15, 2026429 upvotes

View on arXiv Read PDF

AI summary

TL;DR

LimiX-2 is a structured-data model based on Contextual Mechanism Networks that learns joint data distributions conditioned on context, improving in-context learning by modeling underlying data generation mechanisms. It uses synthetic data from structural causal models for pretraining and achieves state-of-the-art performance on tabular benchmarks while enabling causal structure recovery. The approach shifts focus from predicting outcomes to understanding generative mechanisms.

Why It Matters

This work advances structured-data AI by enabling models to learn and reason about causal mechanisms, improving both predictive accuracy and interpretability. The ability to recover causal skeletons from data supports trustworthy decision-making in domains like healthcare and finance. It represents a step toward general-purpose intelligence for tabular data.

Key Findings

LimiX-2 uses a joint modeling objective p(x, y | D_context) instead of traditional target prediction, enhancing in-context learning through mechanism-aware representation.
Pretraining on synthetic data from diverse structural causal models enables robust generalization across varied data structures and generation processes.
The model demonstrates superior performance on TabArena, TALENT, and BCCO benchmarks and accurately recovers causal skeletons via feature attention patterns.

Topics

contextual mechanism networks

structured data

causal inference

in-context learning

transformer

Thank you to arXiv for use of its open access interoperability. arxiv.org. aibrief.ai is not affiliated with or endorsed by arXiv.

LimiX-2: A Contextual Mechanism Network Towards General Structured-Data Intelligence

arXiv: 2609.17488Published: September 15, 2026429 upvotes

TL;DR

Why It Matters

Key Findings

LimiX-2 uses a joint modeling objective p(x, y | D_context) instead of traditional target prediction, enhancing in-context learning through mechanism-aware representation.

Pretraining on synthetic data from diverse structural causal models enables robust generalization across varied data structures and generation processes.

The model demonstrates superior performance on TabArena, TALENT, and BCCO benchmarks and accurately recovers causal skeletons via feature attention patterns.