Sparse Gaussian Conditional Random Fields on Top of Recurrent Neural Networks

Authors: Xishun Wang, Minjie Zhang, Fenghui Ren

AAAI 2018 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental Co R is evaluated by both synthetic data and real-world data, and it shows a significant improvement in performance over state-of-the-art methods.
Researcher Affiliation Academia Xishun Wang, Minjie Zhang, Fenghui Ren School of Computing and Information Technology, University of Wollongong, 2522, NSW, Australia. xw357@uowmail.edu.au; {minjie, fren}@uow.edu.au
Pseudocode Yes Algorithm 1 Alternative training of Co R
Open Source Code No The paper mentions using Theano and Lasagne for implementation, and provides a link to the NPower Forecasting Challenge data and results, but does not explicitly state that the source code for their proposed Co R model is publicly available or provide a link to it.
Open Datasets Yes We apply Co R to an electricity demand prediction problem, which is a competition called NPower Forecasting Challenge 2016. This competition adopted a rolling forecasting mode to simulate the real-world scenario. [...] 1https://www.npowerjobs.com/graduates/forecastingchallenge. Data are publicly available.
Dataset Splits No In each evaluation, random 80% samples are used for training, while the rest are for testing.
Hardware Specification Yes Evaluations are conducted on a server with 8 CPUs and 64 GB Memory.
Software Dependencies No In implementations, we use Theano (Theano Development Team 2016) and Lasagne (Dieleman et al. 2015) for deep neural network.
Experiment Setup Yes The synthetic data are generated as follows. The dimension of feature D is fixed as 10, while the time step T can be varied in {10, 20, 40}, and number of samples N can be varied in {1000, 2000, 4000, 8000}.