A Causal DAG Prior for Synthetic Time-Series Classification Datasets

2026-06-19

AI Summary

Existing methods struggle to generate synthetic multivariate, multi-class time series classification (TSC) data that exhibit cross-modal causal structures. This work proposes a causal directed acyclic graph (DAG)-based synthetic prior that enables the generation of labeled TSC datasets by randomly sampling DAGs over tabular attributes and time series nodes, thereby capturing causal dependencies across channels, time steps, and class labels. The resulting synthetic data are used to fine-tune TabPFN v2.5, marking the first approach capable of jointly modeling multimodal causal structure and temporal dynamics, thus addressing a critical gap in synthetic TSC data generation. Evaluated on 75 UCR/UEA datasets, the method significantly outperforms both the original TabPFN and an ablation variant restricted to tabular modality alone (ROC-AUC, Wilcoxon p = 3.0 × 10⁻⁸).

Abstract

A Prior-data fitted Network learns the posterior predictive induced by its training prior; bringing this paradigm to multivariate time-series classification therefore calls for a synthetic generator that produces complete labelled datasets with temporal structure. We introduce a causal prior that synthesizes each dataset from a randomly sampled DAG over typed nodes across two modalities (tabular attributes and time series), natively producing multivariate, multi-class TSC datasets with cross-modal causal structure across channels, timesteps and labels, a regime not addressed by existing synthetic priors. To validate the prior, we finetune TabPFN v2.5 with minimal adaptations and evaluate on 75 UCR/UEA datasets within TabPFN's operating regime. Finetuning on our generator significantly outperforms both the unmodified upstream model and a tabular-only ablation of the same prior (Wilcoxon signed-rank $p=3.0\times 10^{-8}$ on ROC-AUC), isolating the contribution of the cross-modal temporal structure.

Problem

• synthetic time-series
• causal structure
• multivariate classification
• cross-modal
• DAG prior

Innovation

• causal DAG
• synthetic time-series
• multivariate TSC
• cross-modal structure
• prior-data fitted network