Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

KarmaTS: A Universal Simulation Platform for Multivariate Time Series with Functional Causal Dynamics

Haixin Li, Yanke Li, Diego Paez-Granados

TL;DR

KarmaTS introduces a universal, mixed-initiative platform for simulating multivariate time series via discrete-time structural causal processes, integrating expert knowledge with algorithmic causal discovery to produce realistic, configurable DSCPs. The framework supports lagged and contemporaneous edges, mixed variable types, and modular edge functionals (templates or neural models), enabling targeted data synthesis, causal interventions, and robustness benchmarking. A privacy-aware fMRI synthesis example demonstrates how expert graphs and learned functionals yield realistic synthetic data while preserving sensitive information. Comprehensive benchmarking analyses compare multiple causal-discovery methods across diverse configurations, revealing context-dependent strengths and aligning with prior benchmarks, thereby providing a valuable tool for method development and evaluation in time-series causality. The work highlights KarmaTS’s potential to accelerate causal discovery research by offering realistic, customizable synthetic data and a principled human-in-the-loop workflow.

Abstract

We introduce KarmaTS, an interactive framework for constructing lag-indexed, executable spatiotemporal causal graphical models for multivariate time series (MTS) simulation. Motivated by the challenge of access-restricted physiological data, KarmaTS generates synthetic MTS with known causal dynamics and augments real-world datasets with expert knowledge. The system constructs a discrete-time structural causal process (DSCP) by combining expert knowledge and algorithmic proposals in a mixed-initiative, human-in-the-loop workflow. The resulting DSCP supports simulation and causal interventions, including those under user-specified distribution shifts. KarmaTS handles mixed variable types, contemporaneous and lagged edges, and modular edge functionals ranging from parameterizable templates to neural network models. Together, these features enable flexible validation and benchmarking of causal discovery algorithms through expert-informed simulation.

KarmaTS: A Universal Simulation Platform for Multivariate Time Series with Functional Causal Dynamics

TL;DR

KarmaTS introduces a universal, mixed-initiative platform for simulating multivariate time series via discrete-time structural causal processes, integrating expert knowledge with algorithmic causal discovery to produce realistic, configurable DSCPs. The framework supports lagged and contemporaneous edges, mixed variable types, and modular edge functionals (templates or neural models), enabling targeted data synthesis, causal interventions, and robustness benchmarking. A privacy-aware fMRI synthesis example demonstrates how expert graphs and learned functionals yield realistic synthetic data while preserving sensitive information. Comprehensive benchmarking analyses compare multiple causal-discovery methods across diverse configurations, revealing context-dependent strengths and aligning with prior benchmarks, thereby providing a valuable tool for method development and evaluation in time-series causality. The work highlights KarmaTS’s potential to accelerate causal discovery research by offering realistic, customizable synthetic data and a principled human-in-the-loop workflow.

Abstract

We introduce KarmaTS, an interactive framework for constructing lag-indexed, executable spatiotemporal causal graphical models for multivariate time series (MTS) simulation. Motivated by the challenge of access-restricted physiological data, KarmaTS generates synthetic MTS with known causal dynamics and augments real-world datasets with expert knowledge. The system constructs a discrete-time structural causal process (DSCP) by combining expert knowledge and algorithmic proposals in a mixed-initiative, human-in-the-loop workflow. The resulting DSCP supports simulation and causal interventions, including those under user-specified distribution shifts. KarmaTS handles mixed variable types, contemporaneous and lagged edges, and modular edge functionals ranging from parameterizable templates to neural network models. Together, these features enable flexible validation and benchmarking of causal discovery algorithms through expert-informed simulation.

Paper Structure

This paper contains 73 sections, 21 equations, 16 figures, 3 tables.

Table of Contents

  1. Introduction
  2. Motivation
  3. Contribution
  4. Paper Roadmap
  5. Related Work
  6. Synthetic MTS Generation
  7. Interactive Causal Modeling Interfaces
  8. Benchmarking Platforms
  9. Complementary Roles of KarmaTS and CAUSEME
  10. Causal Discovery in Multivariate Time Series
  11. KarmaTS: An Overview
  12. The Mathematical Model
  13. A Note on the Functionals
  14. Assumptions
  15. Uncertainty Processes
  16. ...and 58 more sections

Figures (16)

  • Figure 1: An illustration of Definition \ref{['def:dscp']}.
  • Figure 2: The user interface for expert knowledge input.
  • Figure 3: An example of a mixed-type graphical model and the simulated MTS.
  • Figure 4: Overview of the human-in-the-loop pipeline with KarmaTS.
  • Figure 5: The outcome of privacy-conscious synthetic data generation. We also provide a quantitative comparison in Appendix \ref{['appendix:fmri more results']}.
  • ...and 11 more figures

Theorems & Definitions (3)

  • definition 1
  • remark 1
  • definition 2