Fast Bayesian Estimation of Point Process Intensity as Function of Covariates

Authors: Hideaki Kim, Taichi Asami, Hiroyuki Toda

NeurIPS 2022 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental We evaluate our algorithm on synthetic and real-world data, and show that it outperforms state-of-the-art methods in terms of predictive accuracy while being substantially faster than a conventional Bayesian method.
Researcher Affiliation Collaboration Hideaki Kim Taichi Asami Hiroyuki Toda NTT Human Informatics Laboratories NTT Corporation {hideaki.kin.cn, taichi.asami.ka}@hco.ntt.co.jp, hirotoda@acm.org
Pseudocode No The paper does not contain any structured pseudocode or algorithm blocks.
Open Source Code Yes Code and data are provided at https://github.com/Hid Kim/APP.
Open Datasets Yes real-world spatial (T R2) data provided by spatstat.data in R (GPL-3) [4]: copper consists of 67 location points of copper ore deposits and 146 line segments representing faults. The covariate of interest is the shortest distance from a given location t to the set of faults (Y R); bei consists of locations of 3605 trees of the species Beilschmiedia pendula and geo-information in a tropical rain forest. The covariates of interest are the terrain elevation and the terrain slope (Y R2); clmfires consists of locations of forest fires in the Castilla-La Mancha region of Spain and the geographical information.
Dataset Splits Yes For each data set, we randomly split the data points into 10 subsets, assigned one to test and the others to training data, and conducted 10-fold cross validations of the predictive performances...
Hardware Specification Yes A Mac Book Pro with 4-core CPU (2.8 GHz Intel Core i7) was used.
Software Dependencies No The paper mentions software like 'spatstat in R' and 'INLA in R' but does not specify their version numbers or the version of R itself, nor does it list other software dependencies with version information.
Experiment Setup Yes We applied to the APPs a multiplicative Gaussian kernel, k(y, y ) = θ0 QDy d=1 e (θd(yd y d))2, where the hyper-parameter θ = (θ0, . . . , θDy) was optimized for each data by maximizing the marginal likelihood (23) through the 25-points grid search.