Machine Learning for Online Algorithm Selection under Censored Feedback

Authors: Alexander Tornede, Viktor Bengs, Eyke Hüllermeier10370-10380

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

Reproducibility Variable Result LLM Response
Research Type Experimental In an extensive experimental evaluation on an adapted version of the ASlib benchmark, we demonstrate that theoretically well-founded methods based on Thompson sampling perform specifically strong and improve in comparison to existing methods.
Researcher Affiliation Academia 1Department of Computer Science, Paderborn University 2Institute for Informatics, LMU Munich
Pseudocode Yes Alg. 1 provides the pseudo code for this revisited Thompson algorithm and a variant inspired by the Buckley-James estimate we discuss in the following.
Open Source Code Yes All code including detailed documentation and the appendix itself can be found on on Git Hub2. https://github.com/alexandertornede/online_as
Open Datasets Yes We base our evaluation on the standard algorithm selection benchmark library ASlib (v4.0) (Bischl et al. 2016)
Dataset Splits No Since ASlib was originally designed for offline AS, we do not use the train/test splits provided by the benchmark, but rather pass each instance one by one to the correspond-ing online approaches, ask them to select an algorithm and return the corresponding feedback.
Hardware Specification Yes All experiments were run on machines featuring Intel Xeon E5-2695v4@2.1GHz CPUs with 16 cores and 64GB RAM
Software Dependencies No The paper does not provide specific software dependencies with version numbers.
Experiment Setup No The corresponding hyperparameter settings used for the experiments can be found in Section F of the appendix and in the repository, parameter sensitivity analyses in Section G.