Material for the AIMS AI for Science summer school tutorial on neural simulation based inference.
At the end of the session, students will be able to:
- Define the core concept of simulation-based inference (SBI) and distinguish implicit-likelihood models from classical likelihood-based approaches.
- Build a normalizing flow by composing affine coupling layers and permutation layers, visualize its performance on the “two moons” problem, and compare it to naive ABC in terms of accuracy and efficiency.
- Use the sbi Python library to approximate posteriors in a neuromuscular simulation problem, evaluating convergence and the impact of neural network hyperparameters.