This repository contains code for calculating complexity- and emergence-related multi-scale measures during Variational Inference (VI, also called approximate Bayesian inference).
We use a hybrid approach—combining numerical and analytical methods—to simulate the evolution of Gaussian parameters during VI. These parameters serve as inputs for calculating:
- Complexity measures such as Integrated Information,
- Emergence-related measures such as Emergence Capacity (based on Partial and Integrated Information Decomposition).
Both types of measures are computed at each point in the evolutionary process.
This work is ongoing and the code is not yet documented and tested for replication and wider use. The project is led by Nadine Spychala in collaboration with Miguel Aguilera.
The main script is mec_var_inf_steady_state_param_sweep.ipynb, which uses functions from mec_var_inf.py located in the src directory. Plotting is handled by mec_var_inf_steady_state_param_sweep_plotting.ipynb.
- Python with Matlab engine support
- Matlab (must be installed locally)
This code is not yet packaged. You'll need to set local-specific directories at the top of both the script and module files.
A publication titled "Exploring the Relation of Variational Inference and Multi-Scale Measures in a Minimal Model" is in preparation. Corresponding updates to this repository will follow.