Probabilistic Center-Surround Model

Bayesian inference based normative model that explains in-vivo surround modulation results in Fu et al. 2026: Statistics of natural scenes shape contextual modulation in the visual cortex.

Setup

Using Docker Compose

Start the development environment:

docker compose up -d --build

Local Installation

Requirements: PyTorch, PyMC, scikit-image, and the insilico-stimuli package.

See Dockerfile for full dependencies.

Main Demonstration

The primary notebook is experiments/exc_driven_model/binary_custom_mapping.ipynb.

Notebook Goals

1. Build the normative model
2. Generate stimuli (maximum excitatory image, MEI + completing surround, MEI + disruptive surround)
3. Perform posterior inference via sampling to obtain neural activity predictions

Visualization

See experiments/exc_driven_model/plots.ipynb to visualize results and reproduce publication figures.

Project Structure

• experiments/ - Main experimental code and notebooks
• probcs/ - Core package modules
• data/ - Datasets and stimuli

Citation

If you use this code, please cite:

Fu et al. 2026. Statistics of natural scenes shape contextual modulation in the visual cortex.