Most of our work is focussed on the automated construction of high-performance effective algorithms for challenging computational problems. Many of our current projects are motivated and driven by two general paradigms, programming by optimisation (PbO) and automated machine learning (AutoML), and a broad range of techniques that enable these paradigms; they involve work on methodological foundations, algorithms, tools and applications to a broad range of well-known problems from academia and industry.
We are well-known for our work on automated algorithm configuration, selection and performance prediction techniques, as well as on stochastic search. In the past, our work has improved the state of the art in solving prominent problems from artificial intelligence, such as propositional satisfiability (SAT), the travelling salesperson problem (TSP), AI planning, supervised machine learning and mixed integer programming (MIP), as well as real-world applications, such as hard- and software verification, flow cytometry, exam timetabling, circuit routing and data centre management (see also our list of publications).
Some Current Projects
Sparkle - a PbO-based problem-solving platform designed to enable the wide-spread and effective use of programming by optimisation (PbO) techniques for improving the state of the art in solving a broad range of prominent AI problems, including SAT and AI Planning.
Grey-Box Algorithm Configuration (with Thomas Bäck @ LIACS) - aims to develop next-generation general-purpose algorithm configuration techniques that limited knowledge about the internal structure of the algorithm to be configured.
Automated Semi-supervised Learning - leverages automated machine learning (AutoML) methods for improving machine learning methods for different tasks involving labelled and unlabelled data.
Next-generation SLS-based SAT Solving (with Shaowei Cai @ Institute of Software, Chinese Academy of Sciences, China) - uses programming by optimisation (PbO) and conceptual advances in stochastic local search (SLS) to improve the state of the art in solving propositional satisfiability problems from various applications.
Multi-objective Algorithm Configuration for Multi-objective Combinatorial Problems (with Aymeric Blot, Laetitia Jourdan and Marie-Éléonore Kessaci @ Inria Lille Nord-Europe, France) - automatically constructs better solvers for multi-objective combinatorial problems using programming by optimisation (PbO) via multi-objective automated algorithm configuration techniques.
Understanding Scientific Progress using Full-text Analysis of Scholarly Publications (with Wout Lamers, Ludo Waltman, Nees Jan van Eck and Paul Wouters @ CWTS) - uses text analysis and machine learning to assess the way in which scientific progress is enabled through published work.
Some Past Projects
SMAC - Sequential Model-based Algorithm Configuration
MO-ParamILS - A Multi-objective Automatic Algorithm Configuration Framework
ParamILS - An Automatic Algorithm Configuration Framework
Hydra - Automatically Configuring Algorithms for Portfolio-Based Selection
Auto-WEKA - Automatic Model Selection and Hyperparameter Optimization in WEKA
AutoFolio - An Automatically Configured Algorithm Selector
SATzilla - Portfolio-based algorithm selection for SAT
AClib - Algorithm Configuration Library
ASlib - Algorithm Selection Library
ESA - Empirical Scaling Analyser
Ablation Analysis - Analysing and Evaluating Parameter Importance
FANOVA - An Efficient Approach for Assessing Hyperparameter Importance
MonoSAT - A SAT Modulo Theory (SMT) Solver for Monotonic Theories over Booleans, Bitvectors and Graphs
SATenstein - Automatically Building Local Search SAT Solvers From Components
UBCSAT - An Implementation and Experimentation Environment for SLS Algorithms for SAT and MAX-SAT