CDE Inference with MDN (Mixture Density Networks)¶
This notebook demonstrates parameter inference using the CDE (Conditional Density Estimation) module with Mixture Density Networks. Unlike SBI-based inference, CDE uses only numpy and autograd, making it lightweight and dependency-free.
Note
This is a placeholder for the full tutorial. The complete interactive notebook will be available in the examples directory.
Overview¶
Key Features:
Lightweight: No PyTorch or SBI dependencies
Fast: Pure numpy implementation with autograd
Flexible: Suitable for various model types
Interpretable: Clear mathematical foundation
When to use CDE vs SBI:
Use CDE when: You want lightweight inference, have limited computational resources, or prefer mathematical transparency
Use SBI when: You need state-of-the-art neural architectures or are working with very high-dimensional problems
Tutorial Content¶
The full tutorial covers:
Generate Synthetic Data: Create nonlinear parameter-observation relationships
Train MDN Estimator: Configure and train the MDN to learn conditional density p(θ|x)
Perform Inference: Use trained model to infer parameters from observations
Visualize Results: Plot posterior distributions and compare with true values
Model Evaluation: Assess quality of posterior approximation
Next Steps¶
Download the complete notebook from the examples directory
Try with different model architectures
Apply to real brain simulation models
Compare with SBI-based approaches