Efficient Bayesian Inference of Sigmoidal Gaussian Cox Processes
Authors: Christian Donner, Manfred Opper
JMLR 2018 | Venue PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | We demonstrate the performance of our method on synthetic data sets and compare with the results of a Monte Carlo sampling method for the model and the variational approximation of Hensman et al. (2015), which we modify to solve the Cox process model with the scaled sigmoid link function. Then we compare our method to the state-of-the-art inference algorithm (Lloyd et al., 2015) on artificial and real data sets with up to 104 observations. |
| Researcher Affiliation | Academia | Christian Donner EMAIL Manfred Opper EMAIL Artificial Intelligence Group Technische Universit at Berlin Berlin, Germany |
| Pseudocode | Yes | Algorithm 1: Variational Bayes algorithm for sigmoidal Gaussian Cox process. |
| Open Source Code | No | The text mentions using the GPflow package and receiving code from other researchers, but there is no explicit statement or link provided by the authors for the open-sourcing of their own methodology described in this paper. |
| Open Datasets | Yes | As an illustrative example we sample a one dimensional Poisson process with the generative model... Next we deal with two real world two dimensional data sets for comparison. The first one is neuronal data, where spiking activity was recorded from a mouse, that was freely moving in an arena (For The Biology Of Memory and Sargolini, 2014; Sargolini et al., 2006). ... As second data set we consider the Porto taxi data set (Moreira-Matias et al., 2013). |
| Dataset Splits | Yes | To evaluate the performance of inference results we are interested in computing the likelihood on test data Dtest, generated from the ground truth... We consider 20000 taxi rides randomly split into training and test set (N = 10017 and N = 9983, respectively). |
| Hardware Specification | No | The paper does not provide specific details about the hardware used for running the experiments. It only discusses runtimes and mentions speed comparisons without specifying CPU, GPU, or memory configurations. |
| Software Dependencies | No | The paper mentions 'ADAM algorithm (Kingma and Ba, 2014)' and 'GPflow package (Matthews et al., 2017)', but does not provide specific version numbers for these or any other software dependencies. |
| Experiment Setup | Yes | For (b) (d) 50 regularly spaced inducing points are used. For (b) (c) 2 103 random integration points are drawn uniformly over the space X, while for (d) X is discretised into the same number of bins... As gradient ascent algorithm we employ the ADAM algorithm (Kingma and Ba, 2014)... Finally, (iii) the value for prior parameters α0 and β0 are chosen such that p(λ) has a mean twice and standard deviation once the intensity one would expect for a homogeneous Poisson Process observing D. The complete variational procedure is outlined in Algorithm 1. |