How to use Edward library for probabilistic modeling with Tensorflow and GPy to study asymptotic connections between Multi-Layer Perceptrons (neural nets) and Gaussian processes?
Computing_With_Infinite_Networks_using_Edwards_and_GPy Numerical verification of the asymptotic connections between Multi-Layer Perceptrons (neural nets) and Gaussian processes. ¶ Authors: Dr. Eren Metin Elçi and Ravinder Kumar ¶ Is there a connection between Multi Layer Perceptrons (neural nets) and Gaussian processes? At least theoretically some supporting arguments were given by Christopher K. I. Williams in https://papers.nips.cc/paper/1197-computing-with-infinite-networks.pdf In this tutorial we use use Edward (library for probabilistic modelling), http://edwardlib.org , which is compatible with Tensorflow. We further use the Gaussian Process implementation provided by GPy. As a side product, we show how to create a custom activation unit for neural networks and how to construct custom Gaussian process kernels. Have fun and if you require the jupyter notebook for this project contact us pyboot@gmail.com! In [1]: