The paper ‘d-MALIBOO: a Bayesian Optimisation framework for dealing with Discrete Variables’, by Roberto Sala, Bruno Guindani and Danilo Ardagna from the Department of Electronics, Information and Bioengineering – Politecnico di Milano with Alessandra Guglielmi from the Department of Mathematics, won the Best Paper Award at the 32nd International Symposium on Modeling, Analysis, and Simulation of Computer and Telecommunication Systems (MASCOTS), held in Krakow, Poland, from 21 to 23 October 2024.
The paper introduces d-MALIBOO, a novel framework that combines Bayesian Optimization (BO) and Machine Learning (ML) to enhance the efficiency of finding optimal configurations in cloud computing environments. It addresses the challenge of optimizing hardware-software setups within a limited evaluation budget, a problem often complicated by the discrete, bounded nature of real-world resource constraints.
Unlike traditional BO, which is designed for continuous domains, d-MALIBOO adapts BO for discrete domains and leverages ML to help identify feasible regions, thus focusing the search more effectively. Experimental results demonstrate that d-MALIBOO significantly outperforms existing methods, particularly in complex scenarios, achieving 2-8 times better results in terms of regret minimization.
