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

LimiX-2, a new tabular foundation model in the LimiX family built on Contextual Mechanism Networks and pretrained with Context-Conditional Masked Modeling, outperforms existing dataset-specific models and tabular foundation models on TabArena, TALENT, and…

Researchers introduced LimiX-2, a new tabular model in the LimiX family developed under the Contextual Mechanism Networks (CMNs) paradigm, with model and data scaling guided by previously established scaling laws. It is pretrained with Context-Conditional Masked Modeling (CCMM) on synthetic datasets generated by structural causal models, spanning diverse graph structures, functional mechanisms, and observation processes. Unlike tabular PFNs centered on p(y | x, D_context), CMNs learn mechanism-oriented joint modeling of p(x, y | D_context), shifting in-context learning from target-centric prediction to mechanism-oriented joint modeling. Evaluations on TabArena, TALENT, and BCCO show LimiX-2 outperforms current dataset-specific models and tabular foundation models. Its feature attention also encodes direct causal relationships, enabling accurate causal skeleton recovery. Both sources (Hugging Face daily papers and arXiv) report consistent details with no discrepancies.

• LimiX-2 is a new tabular model in the LimiX family built on the Contextual Mechanism Networks (CMNs) paradigm.
• CMNs shift in-context learning from target-centric prediction to mechanism-oriented joint modeling of p(x, y | D_context), unlike tabular PFNs centered on p(y | x, D_context).
• Pretraining uses Context-Conditional Masked Modeling (CCMM) on synthetic datasets generated by structural causal models spanning diverse graph structures, functional mechanisms, and observation processes.
• LimiX-2 outperforms dataset-specific models and tabular foundation models on three benchmarks: TabArena, TALENT, and BCCO.
• Its feature attention encodes direct causal relationships, enabling accurate causal skeleton recovery.
• Model and data scaling were guided by previously established scaling laws.