Data Science Meets Optimisation

The Fifth DSO Workshop at IJCAI-22 (id: 50)

Important Dates

  • May 13, 2022 (AOE) May 27, 2022 (AOE): (Extended) deadline for submitting contributions

  • June 3, 2022 May 30, 2022: notification of acceptance

  • July 24: workshop date

Keynote speaker: Diederik M. Roijers (HU University of Applied Science Utrecht & Vrije Universiteit Brussel)

"On the necessity of using multiple objectives in future AI" PDF of Slides

Most real-world problems are not simple. One of the key reasons why they are not, is that the utility that we derive from solutions to these problems depends on multiple (un)desirable properties of these solutions. Humans typically refer to such desirable features as goals, criteria or – as we will call them in this talk - objectives. When objectives are measurable and meaningful, they can form a powerful tool for humans to reason about problems. In AI, objectives are typically formalised by using vector-valued reward functions, where each element in the vector corresponds to an objective.

In this talk, I will argue the following positions: 0) That multiple objectives are essential to many – if not most – real-world problems. 1) That explicitly modelling multiple objectives is essential for explainable AI as well as human-aligned AI. 2) That multiple objectives will help us make AI systems better maintainable. 3) That we should adopt a utility-based approach when dealing with multiple objectives, i.e., that we should derive solution concepts from all that we know about allowed solutions, the way solutions are applied in practice, and how the user derives their utility from these solutions.

Papers and Schedule

Technical talks are 20 mins + 5 mins questions.

09:30 workshop opening

09:40 invited talk: Diederik M. Roijers (HU & VUB), "On the necessity of using multiple objectives in future AI"

10:45 coffee

11:00 paper session 1

Optimizing Fairness in Transport Network Design using Deep Reinforcement Learning (Dimitris Michailidis, Sennay Ghebreab and Fernando P. Santos) Paper

A Survey on Sustainable Surrogate-Based Optimisation (Laurens Bliek) Paper Slides

Learning MAX-SAT Models from Examples using Genetic Algorithms and Knowledge Compilation (Senne Berden, Mohit Kumar, Samuel Kolb and Tias Guns) Paper

Extrapolating Constraint Networks by Symbolic Classification (Steven Prestwich) Paper

12:40 lunch break

14:00 paper session 2

Combinatorial Optimization in School Allocation and the Impact of Strategy (Mayesha Tasnim, Max Baak, Youri Weesie and Sennay Ghebreab) Paper

Dynamic Scenario Reduction for Simulation Based Optimization Under Uncertainty (Noah Schutte, Eghonghon-Aye Eigbe and Kim van den Houten) Paper

Robust Optimization for Integrated Vehicle and Crew Scheduling Based on Uncertainty in the Main Inputs (Liping Ge, Abtin Nourmohammadzadeh, Stefan Voß and Lin Xie) Paper

15:15 coffee

15:30 paper session 3

Dominance-Based Local Search for Neural Architecture Search (Meyssa Zouambi, Julie Jacques and Clarisse Dhaenens) Paper

Correctional Regret for Predict+Optimize with Unknown Objectives and Constraints (Xinyi Hu, Jasper C.H. Lee, Jimmy H.M. Lee and Allen Z. Zhong) Paper

Fair and Optimal Decision Trees: A Dynamic Programming Approach (Jacobus G.M. van der Linden, Mathijs M. de Weerdt and Emir Demirović) Paper

16:45 open discussion: "Opportunities and Pitfalls when Data Science meets Optimisation"

17:30 closing

About the Workshop

Data science and optimization are closely related. On the one hand, many problems in data science can be solved using optimizers, on the other hand optimization problems stated through classical models such as those from mathematical programming cannot be considered independent of historical data. Examples are ample: Machine Learning (ML) often relies on optimization techniques such as linear or integer programming; reasoning systems have been applied to constrained pattern and sequence mining tasks; a parallel development of metaheuristic approaches has taken place in the domains of data mining and machine learning; methods aimed at high level combinatorial optimization have been shown to strongly profit from configuration, algorithm selection and tuning tools building on historical data; ML models can be embedded in combinatorial optimization problems to address hard-to-model systems, or for validation of the ML model itself; “predict, then optimize” scenarios can be dealt with in an integrated fashion to improve considerably the solution quality.

This workshop continues on DSO@IJCAI2021 (online), DSO@IJCAI2020 (online) and the DSO@IJCAI2019 workshop at the International Joint Conference on Artificial Intelligence 2019 in Macao. The DSO workshop is closely related to the DSO working group of The Association of European Operational Research Societies (EURO). In addition to DSO@FAIM 2018, previous related activities include: DSO@IJCAI-ECAI 2018 (Stockholm); stream at EURO 2018 (Valencia); stream at IFORS 2017 (Quebec); workshop at CPAIOR 2017 (Padua); workshop at CEC 2017 (San Sebastian); the foundational workshop (Leuven). More information about the previous editions of DSO is available at https://www.euro-online.org/websites/dso/.

Aim AND Topics

The aim of the workshop is to organize an open discussion and exchange of ideas by researchers from data science, constraint optimization and operations research in order to identify how techniques from these fields can benefit each other. The workshop invites submissions that include but are not limited to the following topics:

  • Applying data science and machine learning methods to solve combinatorial optimization problems, such as algorithm selection based on historical data, speeding up or driving the search process using machine learning including (deep) reinforcement learning, neural combinatorial optimization, and handling uncertainties of prediction models for decision-making.

  • Using optimization algorithms for the development of machine learning models: such as formulating the problem of learning predictive models as MIP, constraint programming or boolean satisfiability (SAT). Tuning machine learning models using search algorithms and meta-heuristics. Learning constraint models from empirical data.

  • Embedding/encoding methods: combining machine learning with combinatorial optimization, model transformations and solver selection, reasoning over machine learning models. Introducing constraints in (hybrid) machine learning models as well as 'predict and optimize' frameworks.

  • Formal analysis of machine learning models via optimization or constraint satisfaction techniques: safety checking and verification via SMT or MIP, generation of adversarial examples via similar combinatorial techniques.

  • Computing explanations for ML model via techniques developed for optimization or constraint reasoning systems.

  • Applications of integrations of techniques of data science and optimization.

Submission

Authors are invited to send a contribution in the in the IJCAI proceedings format, in the form of:

  • Submission of original work up to 6 pages in length (+ references).

  • Submission of work in progress with preliminary results, and position papers, up to 4 pages in length (+ references).

  • Published journal/conference papers in the form of a 2-pages extended abstracts.

Submission should be prepared following the IJCAI formatting instructions at: https://www.ijcai.org/authors_kit.

The review process is single-blind. The programme committee will select the papers to be presented at the workshop according to their suitability to the aims.

Selected contributors will be invited to submit extended articles to a special issue of a journal.

Submissions through: https://easychair.org/conferences/?conf=dsoijcai2022

Organization

The workshop co-chairs are:

  • Tias Guns (KU Leuven, BE) <tias.guns@kuleuven.be>

  • Michele Lombardi (University of Bologna, IT) <michele.lombardi2@unibo.it>

  • Neil Yorke-Smith (TU Delft, NL) <n.yorke-smith@tudelft.nl>

  • Yingqian Zhang (TU Eindhoven, NL) <yqzhang@tue.nl>

PAST EDITIONS

EURO Working Group on DSO: https://www.euro-online.org/websites/dso/

Endorsed by: