Projective Preferential Bayesian Optimization

Authors: Petrus Mikkola, Milica Todorović, Jari Järvi, Patrick Rinke, Samuel Kaski

ICML 2020 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental In this section we demonstrate the effciency of the PPBO method in high-dimensional spaces, and experiment with various acquisition strategies in numerical experiments on simulated functions. (Section 4); In this section we demonstrate the capability of PPBO to correctly and effciently encode user preferences from pro jective preferential feedback. We consider a material science problem... (Section 5)
Researcher Affiliation Academia 1Helsinki Institute for Information Technology HIIT, De partment of Computer Science, Aalto University, Espoo, Fin land 2Department of Applied Physics, Aalto University, Espoo, Finland 3The University of Manchester, UK.
Pseudocode Yes Algorithm 1 Approximate EIn(ξ, x)
Open Source Code Yes Source code is available at https://github.com/Aalto PML/PPBO.
Open Datasets Yes For f we consider four different test functions: Six-hump camel2D, Hartmann6D, Levy10D and Ackley20D.3 (Footnote 3 points to 'https://www.sfu.ca/~ssurjano/optimization.html' which makes these standard benchmark functions openly accessible.)
Dataset Splits No The paper uses benchmark test functions and a user experiment, but does not provide specific details on train/validation/test dataset splits with percentages or counts for reproducibility.
Hardware Specification Yes All experiments of each test function were run on a computing infrastructure of 24x Xeon Gold 6148 2.40GHz cores and 72GB RAM. (Footnote 4, Section 4)
Software Dependencies No The paper mentions 'quantum mechanical atomistic simulation code FHI-aims (Blum et al., 2009)' but does not provide specific version numbers for it or any other software dependencies needed to replicate the experiment.
Experiment Setup Yes We consider a total budget of 100 queries. The ith-initial query corresponds to ξ = ei, that is, to the ith-coordinate projection, and the reference vector x is uniformly random. (Section 4); The total number of queries was 24, of which 6 were initial queries. The ith-initial query corresponded to ξ = ei, that is, to the ith-coordinate projection. The initial values for the reference coordinate vector x were fxed to the same value across all user sessions. For acquisition, we used the expected improvement by projective preferential query. (Section 5)