![[2602.18662] Large Causal Models for Temporal Causal Discovery](https://arxiv.org/static/browse/0.3.4/images/arxiv-logo-fb.png){kind=logo}
[2602.18662] Large Causal Models for Temporal Causal Discovery
Summary
This paper presents Large Causal Models (LCMs) designed for temporal causal discovery, addressing limitations of traditional dataset-specific approaches by enabling multi-dataset pretraining.
Why It Matters
The development of LCMs represents a significant advancement in causal discovery methodologies, particularly for temporal data. By overcoming the constraints of previous models, LCMs can enhance the accuracy and efficiency of causal inference in diverse applications, making them valuable for researchers and practitioners in machine learning and data science.
Key Takeaways
- LCMs provide a scalable framework for temporal causal discovery.
- They outperform traditional models in accuracy and efficiency, especially in out-of-distribution scenarios.
- The integration of synthetic data with realistic datasets enhances model generalization.
- LCMs enable fast, single-pass inference, improving practical applications.
- Extensive experiments validate LCMs' effectiveness across various benchmarks.
Computer Science > Machine Learning arXiv:2602.18662 (cs) [Submitted on 20 Feb 2026] Title:Large Causal Models for Temporal Causal Discovery Authors:Nikolaos Kougioulis, Nikolaos Gkorgkolis, MingXue Wang, Bora Caglayan, Dario Simionato, Andrea Tonon, Ioannis Tsamardinos View a PDF of the paper titled Large Causal Models for Temporal Causal Discovery, by Nikolaos Kougioulis and 6 other authors View PDF HTML (experimental) Abstract:Causal discovery for both cross-sectional and temporal data has traditionally followed a dataset-specific paradigm, where a new model is fitted for each individual dataset. Such an approach limits the potential of multi-dataset pretraining. The concept of large causal models (LCMs) envisions a class of pre-trained neural architectures specifically designed for temporal causal discovery. Prior approaches are constrained to small variable counts, degrade with larger inputs, and rely heavily on synthetic data, limiting generalization. We propose a principled framework for LCMs, combining diverse synthetic generators with realistic time-series datasets, allowing learning at scale. Extensive experiments on synthetic, semi-synthetic and realistic benchmarks show that LCMs scale effectively to higher variable counts and deeper architectures while maintaining strong performance. Trained models achieve competitive or superior accuracy compared to classical and neural baselines, particularly in out-of-distribution settings, while enabling fast, single-pass ...