Manuel López-Ibáñez

Mare Nostrum (Barcelona Supercomputing Center)

Reproducibility in Optimization Research

Mathematical Optimization, Metaheuristics and Evolutionary Computation are mostly empirical fields. However, reproducibility of published results is not always easy or possible. We published a paper describing the main obstacles to reproducibility and suggestions to overcome them. I have also given shorter lectures for various summer schools summarizing the topic and giving practical advice (and a reproducibility checklist) to make your research in optimization more reproducible.

Expensive Combinatorial Optimization Problems

We have proposed the "The Asteroid Routing Problem: A Benchmark for Expensive Black-Box Permutation Optimization".

With Ekhine Irurozki, we have proposed UMM, an estimation-of-distribution algorithm based on mallows models for expensive black-box permutation problems. Our initial results indicate that UMM produces similar results to GP-based approaches, but it less computationally demanding and can tackle larger problems.

With Leslie Pérez Cáceres and Thomas Stützle, we have examined the performance of different ant colony optimization (ACO) algorithms under a strongly limited budget of 1K evaluations. We compared the performance of classical ACO algorithms to an ACO algorithm that makes use of surrogate modeling of the search landscapes (which we call EGACO). We not only show that tuning algorithms for the limited budget case is of uttermost importance, but also that, surprisingly, direct search by the classical ACO algorithms keeps an edge over the EGACO variants using surrogate modeling.

Machine Decision-Makers

The performance assessment of interactive multi-objective (multi-criteria) optimization algorithms is complicated because of the need of a realistic decision-maker that interacts with the algorithm. Together with Joshua Knowles, we have proposed a conceptual framework for the problem of quantitative assessment, based on the definition of machine decision-makers (machine DMs, also called virtual DMs), made somewhat realistic by the incorporation of various non-idealities. The machine DM proposed draws from earlier models of DM biases and inconsistencies in the MCDM literature.

Anytime Behavior of Optimization Algorithms

In joint work with Thomas Stützle, we have proposed to evaluate the anytime behavior of optimization algorithms using the hypervolume measure in order to apply automatic configuration tools like irace.
Together with Jérémie Dubois-Lacoste and Thomas Stützle, we have proposed variants of two-phase local search (TPLS) and Pareto local search (PLS) with much better anytime behavior. These variants can be executed without specifying a termination criterion, and they can be stopped at anytime during their execution, and still return high-quality Pareto front approximations. Hybrid algorithms such as TPLS+PLS are currently (2013) state-of-the-art for several bi-objective combinatorial optimization problems.

Automatic Configuration of Optimization Algorithms

I am one of the two developers, and current maintainer, of the irace software package, which implements the Iterated F-Race procedure for automatic configuration (offline parameter tuning) of optimization algorithms.

Automatic Design of Optimization Algorithms

By using irace and a configurable framework of algorithm components, it is possible to automatically find state-of-the-art algorithm designs never seen before in the literature. We have demonstrated the automatic design of multi-objective ant colony optimization algorithms from the Multi-Objective Ant Colony Optimization Framework.
More recently, we have shown that it is possible to automatically design multi-objective evolutionary algorithms (MOEAs) that outperform the best-known MOEAs from the literature.
A different approach for automatic algorithm design uses a grammar description of the design space. We have shown that irace can outperform previous results that used grammatical evolution for the automatic design of hybrid local search.

Local Search and Local Optima in Multi-Objective Optimization

The properties of local optimal solutions and (Pareto) local optimal sets are not yet well-understood. Together with colleagues Luis Paquete, Arnaud Liefooghe, Bilel Derbel, Sébastien Verel, Hernán E. Aguirre, and Kiyoshi Tanaka, we have done several studies on this topic. See [Liefooghe et al., 2018 (PPSN); Liefooghe et al., 2018 (GECCO); López-Ibáñez et al., 2014].
New Initialisation Techniques for Multi-Objective Local Search: Application to the Bi-objective Permutation Flowshop.

Multi-Objective Optimization

Multi-Objective Ant Colony Optimization Framework is a flexible software framework that is able to instantiate most MOACO algorithms from the literature, and also combine components that were never studied in the literature. Combined with irace, it enables the automatic design of MOACO algorithms.
Sequential Online Archiving of Objective Vectors. Joshua Knowles, Marco Laumanns and I have studied the problem of maintaining a bounded-size archive of (Pareto) nondominated points. We provide software implementing several well-known archivers and benchmark sequences for testing the performance of archiving algorithms.
Graphical tools for exploring the performance of stochastic local search algorithms for biobjective optimization problems based on the concept of the empirical attainment function (EAF). In particular, we consider the visualization of attainment surfaces and differences between the first-order EAFs of two algorithms. This visualization allows to identify certain algorithmic behaviors in a graphical way. Luis Paquete, Thomas Stützle and I have made available software implementing these EAF graphical tools.
The hypervolume indicator is widely used to assess the performance of multi-objective optimisation algorithms. Carlos M. Fonseca, Luis Paquete and I have implemented one of the fastest existing algorithms for calculating the hypervolume indicator.
The evolutionary multiobjective optimization mailing list (emo-list) is dedicated to the discussion of issues related to multiobjective (or multicriteria) optimization and evolutionary computation.

Multi-objective Permutation Flow-shop Scheduling

Together with Jérémie Dubois-Lacoste and Thomas Stützle, we have proposed a hybrid TP+PLS algorithm that is the current (2010) state-of-the-art algorithm for several bi-objective variants of the permutation flow-shop scheduling problem, combining makespan, sum of flowtimes (completion time) and total (weighted) tardiness. We also make available the benchmark sets used and the reference sets obtained.

Travelling Salesman Problem with Time Windows (TSPTW)

Christian Blum and I have collected benchmark instances utilized in the literature for the traveling salesman problem with time windows (TSPTW). These instances are available from different sources, along with instructions on how to modify them to match previous results. However, we provide them all together in a single place, converted into a standard format. Moreover, we provide C++ source code for reading the instance files.
We also published several works (López-Ibáñez et al., 2010, López-Ibáñez et al., 2013) applying Beam-ACO to the TSPTW. Source code is available.

Operational Research applied to Space Exploration

We have proposed the "The Asteroid Routing Problem" as a simplified model of the problem of visiting multiple asteroids while minimizing fuel cost and travel time.

I have given two tutorials at GECCO on the topic of Optimization Challenges at the European Space Agency.

Pump Scheduling in Water Distribution Networks

The extended EPANET Toolkit is a modified version of the EPANET Toolkit that provides additional functions and features for its use in optimisation algorithms without breaking backwards compatibility (under some assumptions). This version has been mainly tested on GNU/Linux, but it should also compile under Win32 environments.
I have also developed a thread-safe variant of EPANET Toolkit to be used by parallel applications.
Instances of the Pump Scheduling Problem, that is, examples of Water Distribution Networks published in the literature and their corresponding EPANET input files.

Water Distribution Network

IRIDIA BibTeX Repository

The IRIDIA BibTeX Repository is a collection of high-quality BibTeX files related to my research. Contributions are welcome via github.