Inverse Stochastic

Codebase for Solving Inverse Problems in Stochastic Self-Organising Systems through Invariant Representations paper. The codebase contains:

Main directories:

• src/models/: Models for reaction-diffusion and Schelling systems.
• src/trainers/: Training and optimization routines, including evolutionary strategies.
• src/vision_models/: Tools for creating embedding representations using CLIP.
• src/visualisations/: Visualization utilities for results and experiments.
• src/scrips_analysis/: Scripts for analyzing and plotting experiment data.
• src/utils/: General utility functions.

---

Installation

Install dependencies using uv and the provided pyproject.toml or uv.lock.

First install uv, then simply type uv sync.

---

Usage

Running Reaction-Diffusion Experiments

To reproduce the reported reaction-diffusion experiment, run:

python experiments/RD_experiment.py

This will optimize parameters for the Gray-Scott reaction-diffusion system to match target patterns, saving results in the experiments_results/RD/ directory.

Running Schelling Model Experiments

To reproduce the reported Schelling segregation model experiment, run:

python experiments/SH_experiment.py

Results will be saved in the experiments_results/SH/ directory.

Running Blastocyst Experiments

Running the blastocyst experiment requires installing Morpheus and setting up the environment as per the instructions on their site, then run:

python experiments/BL_experiment.py

---

Running specific targets

To train for specific target parameters, use the trainer module directly:

• See src/trainers/trainer.py and run it with your desired settings.
• Edit the configuration at the top of the file to set your target system and parameters.

Run:

python src/trainers/trainer.py

Results and outputs will be saved in the specified run folder.