Fast Bayesian Intensity Estimation for the Permanental Process

Authors: Christian J. Walder, Adrian N. Bishop

ICML 2017 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental In section 6 we present some numerical experiments. We compare our new Laplace Bayesian Point Process (LBPP) with two covariances... We obtain orders of magnitude speed improvements over previous work. Figure 4. Mean one standard error performance on the toy problems...
Researcher Affiliation Academia 1Data61, CSIRO, Australia 2The Australian National University 3University of Technology Sydney. Correspondence to: Christian <christian.walder@anu.edu.au>.
Pseudocode No The paper does not contain any structured pseudocode or algorithm blocks.
Open Source Code No The paper does not provide any information about the availability of open-source code for its methodology.
Open Datasets Yes We compared the methods on three real world datasets, coal: 190 points in one temporal dimension, indicating the time of fatal coal mining accidents in the United Kingdom, from 1851 to 1962 (Collins, 2013); redwood: 195 California redwood tree locations from a square sampling region (Ripley, 1977); cav: 138 caveolae locations from a square sampling region of muscle fiber (Davison & Hinkley, 2013).
Dataset Splits No The paper describes train/test splits for real-world data and sampling training sets for toy examples, but it does not explicitly mention a 'validation' split.
Hardware Specification No The paper does not provide specific hardware details (e.g., GPU/CPU models, memory) used for running its experiments.
Software Dependencies No The paper does not specify any software dependencies with version numbers.
Experiment Setup Yes All inference is performed with maximum marginal likelihood, except for KS+EC where we maximise the leave one out metric described in (Lloyd et al., 2015)... Once again we employed the ML-II procedure to determine a and b, fixing m = 2, for the covariance function of subsection 5.1, using the lowest 32 cosine frequencies in each dimension for a total of N = 322 basis functions in the expansion (16). For ease of visualisation we also fixed a = b.