Embedded Bandits for Large-Scale Black-Box Optimization

Authors: Abdullah Al-Dujaili, S. Suresh

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

Reproducibility Variable Result LLM Response
Research Type Experimental Furthermore, numerical experiments were conducted to validate its performance. The results show a clear performance gain over recently proposed random embedding methods for large-scale problems, provided the intrinsic dimensionality is low.
Researcher Affiliation Collaboration Abdullah Al-Dujaili,1 S. Suresh1,2 1 School of Computer Science and Engineering, NTU, Singapore 2 ST Engineering NTU Corporate Lab, Singapore
Pseudocode Yes Algorithm 1 The EMBEDDEDHUNTER Algorithm
Open Source Code Yes The code/data/supplemental materials of this paper will be made available at the project s website: http://ash-aldujaili.github.io/eh-lsopt.
Open Datasets Yes the Ellipsoid, Fletcher Powell, Rosenbrock, and Ackley test functions (Molga and Smutnicki 2005)
Dataset Splits No The paper evaluates algorithms on mathematical test functions rather than traditional datasets with explicit train/validation/test splits. It specifies experiment configurations like the number of function evaluations and dimensionality ranges, but not data partitioning for training and validation sets in a typical machine learning sense.
Hardware Specification No The paper mentions that algorithms were implemented in Python and test functions imported from a Python package but provides no details on the specific hardware used for experiments.
Software Dependencies No The compared algorithms were implemented in Python and the test functions were imported from the Optproblems Python package (Wessing 2016). While Python and the Optproblems package are mentioned, no specific version numbers for either are provided, preventing full reproducibility of the software environment.
Experiment Setup Yes Table 1: Experiments setup. Unless specified above, we set v = 104, n = 104, d = 10, and M = 5. The search space Y for RESOO and EMBEDDEDHUN T E R was set to [ d/η, d/η]d with η = 0.3, SRESOO s Y was set to [ 1, 1]d+1 as suggested in (Qian and Yu 2016; Qian, Hu, and Yu 2016), respectively. hmax was set to the square root of number of function evaluations for each tree. i.e., for RESOO and SRESOO, hmax = v/M; for EMBEDDEDHU N T E R, hmax = v.