LION18 Scope

The 18th Learning and Intelligent OptimizatioN Conference

This meeting, which continues the successful series of LION events (LION 14 in Athens, LION 16 in Milos Island, LION 17 in Nice), is exploring the intersections and uncharted territories between machine learning, artificial intelligence, mathematical programming and algorithms for hard optimization problems.

The main purpose of the event is to bring together experts from these areas to discuss new ideas and methods, challenges and opportunities in various application areas, general trends and specific developments.

The large variety of heuristic algorithms for hard optimization problems raises numerous interesting and challenging issues. Practitioners are confronted with the burden of selecting the most appropriate method, in many cases through an expensive algorithm configuration and parameter tuning process, and subject to a steep learning curve. Scientists seek theoretical insights and demand a sound experimental methodology for evaluating algorithms and assessing strengths and weaknesses. A necessary prerequisite for this effort is a clear separation between the algorithm and the experimenter, who, in too many cases, is "in the loop" as a crucial intelligent learning component. Both issues are related to designing and engineering ways of "learning" about the performance of different techniques, and ways of using past experience about the algorithm behavior to improve performance in the future. Intelligent learning schemes for mining the knowledge obtained from different runs or during a single run can improve the algorithm development and design process and simplify the applications of high-performance optimization methods. Combinations of algorithms can further improve the robustness and performance of the individual components provided that sufficient knowledge of the relationship between problem instance characteristics and algorithm performance is obtained.

Invited talks

Frank Hutter
(Professor for Machine Learning at the University of Freiburg, Germany)

Title: "AI that Builds and Improves AI: Meta-Learning The Next Generation of Learning Methods"

Abstract: Throughout the history of AI, there is a clear pattern that manual elements of AI methods are eventually replaced by better-performing automatically-found ones; for example, deep learning (DL) replaced manual feature engineering with learned representations. The logical next step in representation learning is to also (meta-)learn the best architectures for these representations, as well as the best algorithms and hyperparameters for learning them. In this talk, I will discuss several works along these lines from the field of automated machine learning (AutoML). Specifically, I will discuss the efficiency of AutoML, its relationship to foundation models, its ability to democratize machine learning, and that it can also be extended to optimize various dimensions of trustworthiness (such as algorithmic fairness and robustness). Finally, taking the idea of meta-learning to the extreme, I will deep-dive into a novel approach that learns an entire classification algorithm for small tabular datasets that achieves a new state of the art at the cost of a single forward pass.

Ruth Misener
(Professor in Computational Optimisation, Imperial College London, UK)

Title: "Optimal decision-making problems with trained surrogate models embedded"

Abstract: Several of our recent projects (and complementary projects by other groups worldwide) embed data-driven surrogate models into larger optimal decision-making problems. For example, with the chemicals company BASF, we considered solving inverse problems over trained graph neural networks to design new molecules. This presentation discusses some of the mathematical challenges and practical applications we have explored. We also mention software implementations and close with open challenges in the area.

Matthias Poloczek
(Principal Scientist at Amazon, USA)

Title: "Reliable High-Dimensional Bayesian Optimization for Combinatorial and Mixed Spaces"

Abstract: Impactful applications such as materials discovery, hardware design, AutoML, or portfolio optimization require optimizing high-dimensional black-box functions with mixed and combinatorial input spaces. These functions do not have a closed-form and are evaluated for example by running a complex economic simulation, an experiment in the lab or in a market, or a CFD simulation. A series of recent advances have extended the scope of Bayesian optimization to dozens of variables which allows practitioners to optimize their designs more granularly. However, a closer examination reveals that the state-of-the-art methods suffer from degrading performances or failures.

In this talk I will start with a brief introduction to Bayesian optimization and then discuss the problems that BayesOpt methods encounter in high-dimensional spaces and for mixed input variables. To fill the need for a dependable algorithm for combinatorial and mixed space, I will present the Bounce (Bayesian optimization using increasingly high-dimensional combinatorial and continuous embeddings) algorithm (NeurIPS 2023) and show comprehensive experimental results that demonstrate that Bounce reliably achieves excellent results and outperforms the state-of-the-art methods. Time-permitting, I will also discuss the challenges of productionizing ML-based solutions at scale and how to address them.

Kevin Tierney
(Professor of Decision and Operation Technologies at Bielefeld University, Germany)

Title: "Deep Reinforcement Learning for Vehicle Routing Problems"

Abstract: Learning to automatically construct solutions to vehicle routing problems offers a way to find high-quality solutions to routing problems without needing a human to design an algorithm by hand. The primary goals of this work are to democratize Operations Research (OR), allowing people to solve problems without advanced knowledge in OR, and to automatically adjust algorithms to better solve specific instance sets. However, to date these methods have struggled to beat state-of-the-art human-designed heuristics and generally cannot handle complex side constraints. In this talk, I will discuss the state-of-the-art for learning to route with deep reinforcement learning and provide ideas for overcoming current deficits. Furthermore, I will describe our ongoing work on combining deep learned models with high-level search strategies, such as efficient active search (EAS), simulation-guided beam search (SGBS), and using diversification strategies to improve search performance. I provide experimental results on several routing problems, including the traveling salesperson problem and versions of the capacitated vehicle routing problem, emphasizing the steadily narrowing gap between learned methods and “traditional” heuristics.

You can find the slides of the presentation here.

Mauricio G. C. Resende
(University of Washington, Seattle)

Title: "Random-Key Optimizers (RKO): Problem independent combinatorial optimization"

Abstract: This tutorial introduces Random-Key Optimizers, a problem-independent approach to solving combinatorial optimization problems. Random keys are randomly generated real numbers in the interval (0,1]. A random key vector is a vector of n random keys and corresponds to a point in n-dimensional unit hypercube. Solutions of combinatorial optimization problems can be encoded as random-key vectors. By using a decoder a solution can be retrieved from a random-key vector. For example, by sorting a random key vector, a permutation will result from the indices of the sorted vector. In a TSP these indices could be cities in a tour. In the random-key optimization problem we seek a point in the n-dimensional unit hypercube that optimizes the decoder function. For a TSP, what is the point in the n-dimensional unit hypercube that decoded with the sorting decoder results in the shortest tour on the n cities. We illustrate this concept with three RKOs: a biased random-key genetic algorithm (BRKGA), a dual annealing RKO, and a Continuous GRASP (C-GRASP) RKO.

Proceedings

Papers accepted into the LION18 proceedings were published in Lecture Notes in Computer Science (LNCS).

The proceedings are now available here.

Special sessions

In addition to submissions about general LION themes, we also welcome submissions related to one of our special sessions. The special sessions will be part of the regular conference and are subject to the same peer-review as all other submissions.

Special session 1: ADVANCES AND PERSPECTIVES IN BAYESIAN OPTIMIZATION

Organizer: Antonio Candelieri^1
1University of Milano-Bicocca, Italy

Abstract: Bayesian Optimization (BO) is the most widely adopted learning-and-optimization framework in many real-life applications. The reason underlying its success is that BO is particularly well suited for solving black-box and expensive problems, quite common in crucial sectors such as chemical and material engineering, aerodynamic design, system control, and (Automated) Machine and Deep Learning. The increasing application of BO has required to address new and specific challenges, leading to extensions of the basic framework – aka vanilla BO – from the theoretical and the methodological perspectives. This Special Session will consider both application-driven and theoretical/methodological contributions addressing recent open-challenges and proposing advances and perspectives in BO, such as – but not restricted to: High-Dimensional BO, multi-task and multi-objective BO, multi-fidelity and multiple information sources BO, Safe and Fair BO, cost-aware BO, multiform BO, Transfer Learning for BO, non-Euclidean BO.

Special session 2: Learning and Intelligent Optimization for Physical Systems

Organizer: Konstantinos Chatzilygeroudis¹, Michael Vrahatis^1
1University of Patras, Greece

Abstract: Several critical challenges arise when operating with physical systems contrary to theoretical models, simulated environments, or static datasets. Firstly, reducing the up-time of experimenting with the systems is essential. Experimenting extensively on a physical system might lead to hardware failures that are expensive to replace. Secondly, the algorithm should never produce behaviors that might harm the humans around it or the system itself (e.g., we do not want to break a robot that costs 2M euros). Therefore, to develop effective Machine Learning or Intelligent Optimization methods on physical systems, one has to consider the above challenges during the process of designing the algorithms. Learning and data-driven methods can learn very complex models/controllers and improve over time which is useful when operating with physical systems. However, such methods require a prohibited amount of samples to work reliably, and providing formal guarantees on the obtained solutions is challenging. On the other hand, traditional mathematical optimization is more often used in physical systems since it can operate with no or little data and provide solid theoretical foundations, but it is not easy to make an algorithm that can improve the performance over time. This special session welcomes submissions on "Learning and Intelligent Optimization for Physical Systems", where the goal is to find novel methods that effectively combine data-driven/ML approaches with mathematical optimization to solve tasks on physical systems. Examples are robot learning for control, sensors, embedded systems/mobile phone algorithms, real-time systems/applications, and human-computer interaction.

Program

Important dates

All deadlines are Anywhere on Earth (AoE = UTC-12h).

• April 1, 2024, registration opens
• May 3, 2024, early registration deadline
• May 3, 2024, revision submission closes
• May 15, 2024, conference pre-proceedings
• June 9-13, 2024, conference at Ischia, Italy

Organization

Chairs

General chair

Paola Festa (University of Napoli “Federico II”, Italy)

Steering Committee

Roberto Battiti (University of Trento, Italy - Head of the Steering Committee)  
Francesco Archetti (Consorzio Milano Ricerche, Italy)  
Christian Blum (Spanish National Research Council (CSIC), Spain)  
Mauro Brunato (University of Trento, Italy)  
Carlos A. Coello-Coello (CINVESTAV-IPN, Mexico)  
Clarisse Dhaenens (University of Lille, France)  
Paola Festa (University of Napoli, Italy)  
Martin Charles Golumbic (University of Haifa, Israel)  
Youssef Hamadi (Tempero Tech, France)  
Laetitia Jourdan (University of Lille, France)  
Nikolaos Matsatsinis (Technical University of Crete, Greece)  
Panos Pardalos (University of Florida, USA)  
Mauricio Resende (University of Washington, USA)  
Meinolf Sellmann (InsideOpt, USA)  
Yaroslav Sergeyev (University of Calabria, Italy)  
Dimitris Simos (SBA Research, Austria)  
Thomas Stuetzle (University of Bruxelles, Belgium)  
Kevin Tierney (Bielefeld University, Germany)

Technical Program Committee:

• Carlos Ansòtegui (University of Lleida, Spain)
• Francesco Archetti (Consorzio Milano Ricerche, Italy)
• Annabella Astorino (ICAR-CNR, Italy)
• Hendrik Baier (Eindhoven University of Technology, The Netherlands )
• Roberto Battiti (University of Trento, Italy)
• Laurens Bliek (Eindhoven University of Technology, The Netherlands )
• Christian Blum (Spanish National Research Council (CSIC), Spain)
• Mauro Brunato (University of Trento, Italy)
• Zaharah Bukhsh (Eindhoven University of Technology, The Netherlands )
• Sonia Cafieri (Ecole Nationale de l'Aviation Civile, France)
• Antonio Candelieri (University of Milano Bicocca, Italy)
• Zhiguang Cao (Singapore Management University, Singapore)
• Marco Chiarandini (University of Southern Denmark, Denmark)
• John Chinneck (Carleton University, Canada)
• Konstantinos Chatzilygeroudis (University of Patras, Greece)
• Philippe Codognet (JFLI / Sorbonne Universitè, Japan / France)
• Patrick De Causmaecker (Katholieke Universiteit Leuven, Belgium)
• Renato De Leone (University of Camerino, Italy)
• Clarisse Dhaenens (Université Lille 1 (Polytech Lille, CRIStAL, INRIA), France)
• Luca Di Gaspero (DPIA - University of Udine, Italy)
• Theresa Elbracht (Bielefeld University, Germany)
• Adil Erzin (Sobolev Institute of Mathematics)
• Giovanni Fasano (University Ca'Foscari of Venice, Italy)
• Daniele Ferone (University of Napoli FEDERICO II, Italy)
• Paola Festa (University of Napoli FEDERICO II, Italy)
• Adriana Gabor (Khalifa University, Abu Dhabi)
• Jerome Geyer-Klingeberg (Celones, Germany)
• Isel Grau (Eindhoven University of Technology, The Netherlands )
• Vladimir Grishagin (Nizhni Novgorod State University, Russia)
• Mario Guarracino (ICAR-CNR, Italy)
• Francesca Guerriero (University of Calabria, Italy)
• Ioannis Hatzilygeroudis (University of Patras, Greece)
• Youssef Hamadi (Tempero, France)
• Andre Hottung (Bielefeld University, Germany)
• Laetitia Jourdan (INRIA/LIFL/CNRS, France)
• Marie-Eleonore Kessaci (Université de Lille, France)
• Michael Khachay (Krasovsky Institute of Mathematics and Mechanics, Russia)
• Elias B. Khalil (University of Toronto, Canada)
• Yury Kochetov (Sobolev Institute of Mathematics, Russia)
• Ilias Kotsireas (Wilfrid Laurier University, Waterloo, Canada)
• Dmitri Kvasov (DIMES, University of Calabria, Italy)
• Dario Landa-Silva (University of Nottingham, United Kingdom)
• Hoai An Le Thi (Université de Lorraine, France)
• Daniela Lera (University of Cagliari, Italy)
• Yuri Malitsky (FactSet, USA)
• Vittorio Maniezzo (University of Bologna, Italy)
• Silvano Martello (University of Bologna, Italy)
• Yannis Marinakis (Technical University of Crete, Greece)
• Nikolaos Matsatsinis (Technical University of Crete, Greece)
• Laurent Moalic (University of Haute-Alsace - IRIMAS, France)
• Hossein Moosaei (Jan Evangelista Purkyně University, Czech Republic)
• Tatsushi Nishi (Osaka University, Japan)
• Panos Pardalos (University of Florida, USA)
• Axel Parmentier (Ecole Nationale des Ponts et Chaussées, France)
• Konstantinos Parsopoulos (University of Ioannina, Greece)
• Vincenzo Piuri (Universita' degli Studi of Milano, Italy)
• Oleg Prokopyev (University of Pittsburgh, USA)
• Michael Römer (Bielefeld University, Germany)
• Massimo Roma (SAPIENZA Universita' of Roma, Italy)
• Valeria Ruggiero (University of Ferrara, Italy)
• Frédéric Saubion (University of Angers, France)
• Andrea Schaerf (University of Udine , Italy)
• Elias Schede (Bielefeld University, Germany)
• Marc Schoenauer (INRIA Saclay Île-de-France, France)
• Meinolf Sellmann (InsideOpt, USA)
• Marc Sevaux (Lab-STICC, Université de Bretagne-Sud, France)
• Paul Shaw (IBM, France)
• Dimitris Simos (SBA Research, Austria)
• Thomas Stützle (Université Libre de Bruxelles (ULB), Belgium)
• Tatiana Tchemisova (University of Aveiro, Portugal)
• Kevin Tierney (Bielefeld University, Germany)
• Gerardo Toraldo (Università della Campania “Luigi Vanvitelli”, Italy)
• Paolo Turrini (University of Warwick, UK)
• Michael Vrahatis (University of Patras, Greece)
• Om Prakash Vyas (Indian Institute of Information Technology , India)
• Ranjana Vyas (Indian Institute of Information Technology , India)
• Dimitri Weiß (Bielefeld University, Germany)
• Daniel Wetzel (Bielefeld University, Germany)
• David Winkelmann (Bielefeld University, Germany)
• Dachuan Xu (Beijing University of Technology, Chine)
• Qingfu Zhang (University of Essex & City U of HK, Hong Kong)
• Anatoly Zhigljavsky (Cardiff University, United Kingdom)
• Antanas Zilinskas (Vilnius University, Lithuania)

Local Organizing Committee:

• Maurizio Bruglieri
• Ciriaco D'Ambrosio
• Daniele Ferone
• Paola Festa (Chair)
• Giusy Macrina
• Tommaso Pastore
• Ornella Pisacane

Location, travel, accommodation

Ischia Island, Naples, Italy

Ischia is one of the wonderful islands in the Gulf of Naples, having volcanic origin and known and appreciated all around the world for its diversified landscape, natural beauty, and thermal water.

Its wonderful thermal hot springs have been used for wellness and therapeutic treatments since the VII century b.C. On Ischia, there are many nice beaches that invite the visitor to take a swim.

Conference Hotel

Hotel Continental Terme has an architecture expressed in the Mediterranean style buildings surrounded by the lush greenery of a park.

The Conference Centre counts 12 comfortable modular meeting rooms hosting from 15 to 300 seats.

It offers to participants the opportunity to combine work with a short holiday of sun, sea and wellness with the added value that only an enchanting place like Ischia can give.

Contacts