Nonmyopic Multifidelity Acitve Search

Authors: Quan Nguyen, Arghavan Modiri, Roman Garnett

ICML 2021 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental We evaluate the performance of our solution on real-world datasets and demonstrate significantly better performance than natural benchmarks.
Researcher Affiliation Academia 1Washington University in St. Louis, MO, USA 2University of Toronto, Toronto, Canada.
Pseudocode Yes We give the pseudocode for the policy in the appendix.
Open Source Code Yes Matlab implementations of our policies are available at: https://github.com/KrisNguyen135/multifidelity-active-search .
Open Datasets Yes Here we used the first 50 proteins from the Binding DB database (Liu et al., 2007) described by Jiang et al. (2017). A set of 100 000 compounds sampled from the ZINC database (Sterling & Irwin, 2015) served as a shared negative set. ...This dataset comprises 106 810 alloys from the materials literature (Kawazoe et al., 1997; Ward et al., 2016)
Dataset Splits No The paper does not explicitly provide training/validation/test dataset splits with percentages or sample counts. It describes an iterative active search process within a budget.
Hardware Specification No The paper does not provide specific hardware details (e.g., GPU/CPU models, memory) used for running the experiments.
Software Dependencies No The paper states 'Matlab implementations of our policies', but does not provide specific version numbers for Matlab or any other software dependencies.
Experiment Setup Yes We set θ {0.1, 0.3}. We set k, the number of L queries that are made for each H query, to be either 2 or 5, and set the budget on H to be 300. ... In our experiments, we set u = s = 500 for MF ENS. ... We set β = 0.01 for L queries and β = 0.001 for H queries, as suggested in the same work.