Expected Hypervolume Improvement Is a Particular Hypervolume Improvement

Authors: Jingda Deng, Jianyong Sun, Qingfu Zhang, Hui Li

AAAI 2025 | Venue PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental We conduct a number of experiments to validate the correctness and efficiency of our formulation. The results suggest that our implementation outperforms the existing fastest implementation for EHVI and q EHVI when q 4. For q > 4, our implementation stands out when the number of objectives are large enough.
Researcher Affiliation Academia 1School of Mathematics and Statistics, Xi an Jiaotong University 2Department of Computer Science, City University of Hong Kong 3The City University of Hong Kong Shenzhen Research Institute
Pseudocode No The paper describes theoretical formulations, theorems, and proofs but does not include any explicitly labeled pseudocode or algorithm blocks with structured steps.
Open Source Code Yes Our codes and the supplementary materials are available at https://github.com/Ksrma/EHVI-HVI.
Open Datasets No Following researches in the field of EHVI computation (Yang et al. 2019a), we generate artificial random point sets of different sizes n as the point set A in the q EHVI (q 1) problem. Given a set of evaluated solutions D = {x(i), f(x(i))}K i=1 randomly sampled by Bo Torch, a multitask GP model is trained.
Dataset Splits No The paper mentions generating "artificial random point sets" and "randomly sampled" data, and for experiments, they generate "b different batches of test samples". However, it does not specify traditional training, validation, and test splits for a fixed or publicly available dataset.
Hardware Specification Yes The platform is the Bo Torch platform (Balandat et al. 2020) on 64-bit Windows with Intel 2.60GHz Core i5-13500H processor and 16GB of RAM.
Software Dependencies No The paper mentions using "Bo Torch" and "scipy.stats.multivariate normal class" but does not provide specific version numbers for these software components.
Experiment Setup Yes The experimental settings of the number of objectives m, the number of points n in A, the batch size q, and the number of batches b are presented in Table 2. For each group of (m, n, q), we generate b different batches of test samples {x(i)}q i=1. Table 2: m = 2, 3, 4, 5, 6, 7, 8; n = 10, 20, 40, 80; q = 1, 2, 4, 6; b = 51.