Abstract
We define an evolving in time Bayesian neural network called a Hidden Markov neural network. The weights of a feedforward neural network are modelled with the hidden states of a Hidden Markov model, whose observed process is given by the available data. A filtering algorithm is used to learn a variational approximation to the evolving in time posterior over the weights. Training is pursued through a sequential version of Bayes by Backprop Blundell et al. 2015, which is enriched with a stronger regularization technique called variational DropConnect. The experiments test variational DropConnect on MNIST and display the performance of Hidden Markov neural networks on time series.
Original language  English 

Publication status  Submitted  20 Jun 2020 
Fingerprint
Dive into the research topics of 'Dynamic Bayesian Neural Networks'. Together they form a unique fingerprint.Student Theses

Highdimensional hidden Markov models: methodology, computational issues, solutions and applications
Author: Rimella, L., 28 Sep 2021Supervisor: Whiteley, N. (Supervisor)
Student thesis: Doctoral Thesis › Doctor of Philosophy (PhD)
File