Near-Optimal Active Learning of Multi-Output Gaussian Processes

Authors: Yehong Zhang, Trong Nghia Hoang, Kian Hsiang Low, Mohan Kankanhalli

AAAI 2016 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental Empirical evaluation on real-world datasets shows that our proposed approach outperforms existing algorithms for active learning of MOGP and single-output GP models.
Researcher Affiliation Academia Department of Computer Science , Interactive Digital Media Institute National University of Singapore, Republic of Singapore {yehong, lowkh, mohan}@comp.nus.edu.sg , idmhtn@nus.edu.sg
Pseudocode No The paper describes the approximation algorithm using mathematical equations (9) but does not present it in a pseudocode block or algorithm box format.
Open Source Code No The paper does not provide any statement or link indicating that the source code for the described methodology is publicly available.
Open Datasets Yes Jura dataset (Goovaerts 1997) contains concentrations of 7 heavy metals collected at 359 locations in a Swiss Jura region; (b) Gilgai dataset (Webster 1977) contains electrical conductivity and chloride content generated from a line transect survey of 365 locations of Gilgai territory in New South Wales, Australia; and (c) indoor environmental quality (IEQ) dataset (Bodik et al. 2004) contains temperature ( F) and light (Lux) readings taken by 43 temperature sensors and 41 light sensors deployed in the Intel Berkeley Research lab.
Dataset Splits No The paper states that observations are 'randomly selected to form the test set T' but does not provide specific details on training, validation, or test splits, nor does it mention a validation set.
Hardware Specification No The paper does not provide any specific details about the hardware (e.g., GPU models, CPU types, memory) used to run the experiments.
Software Dependencies No The paper mentions software tools like 'k-means' and 'MOGP and single-output GP models' but does not specify any version numbers for these or other software dependencies.
Experiment Setup Yes For all experiments, k-means is used to select inducing locations U by clustering all possible locations available to be selected for observation into |U| clusters such that each cluster center corresponds to an element of U. The hyper-parameters (i.e., σ2 si, σ2 ni, P0 and Pi for i = 1, . . . , M) of MOGP and single-output GP models are learned using the data via maximum likelihood estimation ( Alvarez and Lawrence 2011). For example, for Jura dataset, |U| = 50, 100, 200 are used.