arXiv:2602.01135v3 Announce Type: replace Abstract: Autoregressive models trained via next-token prediction implicitly learn the conditional independence structure of their data-generating process. We exploit this observation to perform scalable causal discovery from a single observed sequence of discrete events -- without any task-specific retraining. Such single-stream settings arise naturally in vehicle diagnostics, manufacturing systems, and patient trajectories, yet they remain largely unsolved: the absence of repeated samples, massive event vocabularies, and long-range temporal dependenc
The increasing sophistication and widespread adoption of autoregressive models, alongside the growing need for causal inference in complex domains, motivates this exploitation of their inherent capabilities.
This development offers a novel, scalable approach to causal discovery from single data streams, which can unlock new insights and automation opportunities in sectors previously challenged by data scarcity or complexity.
Causal discovery, traditionally reliant on repeated samples or specific experimental setups, now has a powerful new method applicable to 'single-stream settings' using existing autoregressive AI models without bespoke retraining.
- · AI researchers and developers
- · Companies in vehicle diagnostics
- · Manufacturing systems operators
- · Healthcare/patient trajectory analysis
- · Traditional causal inference methods requiring large datasets
- · Solutions reliant on extensive retraining for causal analysis
Existing autoregressive models gain an additional capability: extracting causal graphs implicitly from their learned representations.
This could lead to a proliferation of new AI-driven diagnostic and predictive tools across industrial and medical applications using existing data streams.
Reduced development cycles and costs for causal modeling could accelerate automation and optimization in highly complex, dynamic systems, potentially leading to fully autonomous discovery and intervention systems.
This signal links to a primary source. Continuum Brief monitors and indexes it as part of the live intelligence stream — we do not republish source content.
Read at arXiv cs.LG