Bayesian Nonparametric Spectral Estimation

Authors: Felipe Tobar

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

Reproducibility Variable Result LLM Response
Research Type Experimental Comparison with previous approaches, in particular against Lomb-Scargle, is addressed theoretically and also experimentally in three different scenarios. Code and demo available at github.com/GAMES-UChile. This experimental section contains three parts focusing respectively on: (i) consistency of BNSE in the classical sum-of-sinusoids setting, (ii) robustness of BNSE to overfit and ability to handle non-uniformly sampled noisy observations (heart-rate signal), and (iii) exploiting the functional form of the PSD estimate of BNSE to find periodicities (astronomical signal).
Researcher Affiliation Academia Felipe Tobar Universidad de Chile ftobar@dim.uchile.cl
Pseudocode No No. The paper describes the proposed model and methods mathematically and textually but does not include any structured pseudocode or algorithm blocks.
Open Source Code Yes Code and demo available at github.com/GAMES-UChile.
Open Datasets Yes We next considered two heart-rate signals from http://ecg.mit.edu/time-series/. Lastly, we considered the sunspots dataset, an astronomical time series that is known to have a period of approximately 11 years...
Dataset Splits No No. While the paper mentions using
Hardware Specification No No. The paper does not provide specific hardware details such as exact GPU/CPU models, processor types, or memory amounts used for running the experiments. It mentions using 'GPflow' which implies computational resources, but no specifications are given.
Software Dependencies No No. The paper mentions using
Experiment Setup Yes The window parameter was set to α = 1/(2 502) for an observation neighbourhood much wider than the support of the observations, and we chose an SM kernel with rather permissive hyperparameters: a rate γ = 1/(2 0.052) and θ = 0 for a prior over frequencies virtually uninformative. We implemented BNSE with a lengthscale equal to one and θ = 0 for a broad prior over frequencies, and α = 10 3 for a wide observation neighbourhood.