Alberto Franzin's talk on Simulated Annealing
Revisiting Simulated Annealing: From a Component-Based Analysis to an Automated Design of SA Algorithms
Simulated Annealing (SA) is one of the oldest metaheuristics and has been adapted to solve many combinatorial optimization problems. Over the years, many authors have proposed both general and
problem-specific improvements and variants of SA. We propose to accumulate this knowledge into automatically configurable, algorithmic frameworks so that for new applications that wealth of alternative algorithmic components is directly available for the algorithm designer without further manual intervention.
Here, we describe SA as an ensemble of algorithmic components, and describe SA variants from the literature within these components. We show the advantages of our proposal by (i) implementing existing algorithmic components of variants of SA, (ii) studying SA algorithms proposed in the literature, (iii) improving SA performance by automatically designing new state-of-the-art SA implementations and (iv) studying the role and impact of the algorithmic components based on experimental data.
Our experiments consider three common combinatorial optimization problems, the quadratic assignment problem and two variants of the permutation flow shop problem.
Alberto Franzin is Ph.D. candidate at the IRIDIA lab at the Universitè Libre de Bruxelles, Belgium, under the supervision of prof. Thomas Stuetzle. He is working on automatic algorithm design and configuration, with a focus on heuristics.
Formerly, he was research collaborator at the Department of Information Engineering of the University of Padova under the supervision of Prof. Barbara di Camillo and Dr. Francesco Sambo, developing the bnstruct (cran, repo) R package for learning Bayesian Networks (also) in presence of missing data.
He is author or co-author of 6 papers on scientific journals and international conference proceedings.