ALD-Lab/SURD
Synergistic-Unique-Redundant Decomposition of causality from time series (Nature Communications, 2024)
What it solves
SURD addresses the challenge of understanding complex causal relationships in time-series data. It decomposes the information that past events provide about a future target variable into three distinct categories: redundant (shared information), unique (information from a single source), and synergistic (information created by the interaction of multiple sources). It also quantifies the "causality leak," which represents the portion of the future that remains unexplained by the observed variables.
How it works
The method analyzes the joint probability distribution of past source events and future target events. By default, it uses histograms to estimate these distributions, though it also supports neural-network-based estimation for scenarios with many variables or limited data. The surd.states subpackage extends this by resolving these causal contributions down to the specific states of the variables, allowing researchers to see not just that a variable is causal, but in which states that causality occurs.
Who it’s for
This tool is designed for researchers and scientists working with time-series data from simulations or experiments—such as turbulence studies, population dynamics, or complex system modeling—who need to move beyond simple correlation to understand the specific structure of causal interactions.
Highlights
- Decomposes causality into redundant, unique, and synergistic components.
- Quantifies the causality leak to measure unexplained information.
- Supports both histogram-based and neural-network-based estimation for flexibility.
- Includes state-resolved analysis to identify specific causal states.
- Compatible with multi-variable systems, various time lags, and instantaneous links.
- Proven methodology published in Nature Communications and Communications Physics.
Related
- Project
- Project
- Project
- Project