Welcome to CONNECT’s documentation!

COsmological Neural Network Emulator of CLASS using TensorFlow

Overview

CONNECT is a framework for emulating cosmological parameters using neural networks. This includes both sampling of training data and training of the actual networks using the TensorFlow library from Google. CONNECT is designed to aid in cosmological parameter inference by immensely speeding up the process. This is achieved by substituting the cosmological Einstein-Boltzmann solver codes, needed for every evaluation of the likelihood, with a neural network with a \(10^2\) to \(10^3\) times faster evaluation time.

Citation

Please cite the paper [arXiv:2205.15726] if using CONNECT for publications.

Contents:

• Installation and setup
  • Manual setup
• Usage
  • Creating a neural network
  • Using a trained neural network for MCMC
• Example of workflow - ΛCDM
  • Useful commands for monitoring the iterative sampling
• Support
• CosmoSlider