Time series forecasting based on deep architectures has been gaining popularity in recent years due to their ability to model complex non-lineartemporal dynamics . The recurrent neural network is one such model capable of handling variable-length input and output . In our model design, the transition function of therecurrent neural network, which determines the evolution of the hidden states, is stochastic rather than deterministic as in a regular recurrent neuralnetwork . We test our model on a wide range of data sets from finance to healthcare . Results show that the stochastically recurrentneural network consistently outperforms its deterministic counterpart. We test it on a range ofdatasets from Finance to healthcare; results show that it consistently outranks its deterrant counterpart. Our model can be easily integrated into any deeparchitecture for sequential modelling. It is available to download and use in print and email your favourite from iReport

Author(s) : Zexuan Yin, Paolo Barucca

Links : PDF - Abstract

Code :

https://github.com/alsoj/Recommenders-movielens


Coursera

Keywords : model - network - neural - recurrent - counterpart -

Leave a Reply

Your email address will not be published. Required fields are marked *