Learnijoy
ToolAI ResearchAI Engineering & DevTools

DoTime Generates Synthetic Benchmarks for Causal Time Series Inference

Dennis Thumm, Billy Tim Anthony, Ying Chen· July 31, 2026 View original

Key takeaways

  • DoTime is a new, open-source tool for generating synthetic interventional and counterfactual time series benchmarks.
  • It addresses the critical need for robust evaluation datasets in causal AI for time series.
  • The generator supports advanced features like continuous-time interventions and non-stationary dynamics.
  • Initial findings suggest interventional training improves accuracy over observational models.

Who benefits

HealthcarePolicy & GovernmentClimate ScienceFinanceManufacturing

Summary

Researchers introduce DoTime, an open-source, scalable generator for multivariate temporal structural causal models (TSCMs) that includes continuous-time interventions and counterfactual sampling. This new tool addresses the lack of robust benchmarks for interventional and counterfactual causal inference in time series data.

Causal inference in time series, particularly for interventional and counterfactual scenarios, has been hampered by a scarcity of suitable benchmarks. Existing datasets are often observational, small, or highly domain-specific, making it difficult to rigorously test and compare new causal models. This gap is significant for critical applications in fields like healthcare, policy evaluation, and climate science, where understanding the impact of interventions is paramount. To address this, a new benchmark generator called DoTime has been released as a PyPI package. DoTime creates multivariate temporal structural causal models (TSCMs) and uniquely supports continuous-time intervention windows, counterfactual sampling with positivity guards, and regime-switching SCMs. It also allows for non-stationary dynamics and deterministic intervention profiles, providing a more realistic and challenging environment for causal inference research. The tool comes with four frozen evaluation suites, including a large training-scale snapshot and eight named identification structures, each with exact ground truth for interventional and counterfactual trajectories. Initial experiments using DoTime demonstrate that interventional training provides a measurable advantage in direction-accuracy over purely observational models, validating the generator's utility for advancing causal AI.

Why it matters

Professionals developing or deploying causal AI models for time series analysis need robust benchmarks to validate their methods. DoTime provides a flexible and scalable solution for generating diverse, realistic causal time series data with ground truth, crucial for advancing model development and reliability.

How to implement this in your domain

  1. 1Install the dotime PyPI package to access the synthetic benchmark generator.
  2. 2Utilize DoTime to create custom multivariate temporal structural causal models with specific intervention profiles for research or model testing.
  3. 3Leverage the provided evaluation suites and reference baselines to compare the performance of new causal inference algorithms.
  4. 4Design experiments to test the impact of interventional training on model accuracy and robustness using DoTime's capabilities.

Original post by Dennis Thumm, Billy Tim Anthony, Ying Chen

"arXiv:2607.27263v1 Announce Type: new Abstract: Most benchmarks for causal inference over time series are observational, small, or domain-specific, leaving interventional and counterfactual estimation under-served exactly where it matters most, such as in healthcare, policy evalu…"

View on X

Originally posted by Dennis Thumm, Billy Tim Anthony, Ying Chen on X · view source

Want to go deeper?

Turn these trends into skills with Learnijoy's hands-on AI & tech courses.

Explore courses

More in AI Research

AI ResearchAI Engineering & DevTools

New Framework Improves Partial Multi-View Clustering Performance.

DAS-PMVC is a novel framework for partial multi-view clustering that addresses view asymmetry and irrelevant samples by leveraging dual alignment and structure enhancement. It uses anchor graph structure alignment, structure-enhanced feature learning, and a dual alignment strategy to achieve superior clustering performance on various datasets.

Shubin Ma, Liang Zhao, Chuanye He, Zhenjiao Liu, Liang Zou, Lin Yuanbo Wu, Yu ShaoJul 31, 2026
AI Engineering & DevToolsAI Research

Dual Teachers Improve Adversarial Robustness and Accuracy.

This work extends Information Bottleneck Distillation (IBD) by introducing a "clean teacher" alongside a robust teacher to improve the robustness/accuracy tradeoff against adversarial attacks. The proposed method transfers features from both teachers to a student model, achieving better clean accuracy while maintaining adversarial robustness, outperforming original IBD and competing with state-of-the-art approaches.

Vincent Ryusuke Takahashi, Yoshinari Takeishi, Jun'ichi Takeuchi, Kave SalamatianJul 31, 2026
AI Engineering & DevToolsAI Research

Dynamic Batch Sizes Improve Large Language Model Training Efficiency.

This paper proposes a new approach to deep learning dynamics, deriving joint scaling laws for loss based on both learning rate and batch size schedules. It introduces an optimal dynamic batch size schedule that consistently outperforms static batch size baselines, highlighting its importance for large language model training.

Jiaxiang Li, Zhiqi Bu, Shiyun XuJul 31, 2026