Title Promoter Affiliations Abstract "Declarative Languages and Artificial Intelligence." "Floris Geerts" "Technische universität Darmstadt, Katholieke Universiteit Leuven, Polytechnique Montréal, Catholic University of Chile, Hasselt University, University of Porto, Imperial College London, University of Southern Denmark, University of Namur, University of Copenhagen, University of Hong Kong, Swiss Federal Institute of Technology Lausanne, University of Calabria, Buenos Aires Institute of Technology, University of Mons, University of Kentucky, University of Technology Eindhoven, University of Bayreuth, Simon Fraser University, University of Wisconsin-Madison, University of Edinburgh, Indiana University, TU Dortmund University, University of Bologna, Vrije Universiteit Brussel, ADReM Data Lab (ADReM)" "A network to foster the cooperation between research groups with an interest in the use of declarative methods, promoting international cooperation and stimulating the Flemish groups in maintaining the high quality of their research." "Declarative modeling for machine learning and data mining." "Luc De Raedt" "Declarative Languages and Artificial Intelligence (DTAI)" "Declarative models specify what conditions need to satisfied in order to obtain a solution to a specific problem. Declarative models contrast with the traditional procedural approach which specify how such solutions must be computed. A declarative modeling paradigm will be developed and applied to the areas of machine learning, data mining and experimentation. The declarative modeling paradigm that we will pursue consists of three key components: A modeling language (M language) which is a high level declarative language for specifying the relevant domain knowledge, independent of a particular task (M component). A solver which accepts input in a more low-level task-oriented language (S language) and performs a particular computational task (S component). A programming platform in which a user employs a general purpose language to solve a specific computational task. Today, there exist no general declarative approaches to machine learning, data mining and experimentation. Therefore, contrasting these domains with contemporary approaches to declarative modeling (pursued in knowledge representation and constraint programming) forms an ideal setup for realizing breakthroughs in declarative modeling as well as in machine learning, data mining and experimentation." "An overarching declarative framework for robotics programming" "Nikolaos Tsiogkas" "Declarative Languages and Artificial Intelligence (DTAI), Robotics, Automation and Mechatronics (RAM)" "In the field of robotics programming the same three languages, C, C++ and Python continue to dominate. The aforementioned languages have their benefits: the combo C and C++ for their efficient low-level features; and Python for its low learning curve and its acceptance in non-expert computer scientists and in fields outside of robotics. But more importantly, they have their disadvantages: C and C++ are notoriously error-prone, and the object-oriented features are less mature than in modern object-oriented languages. Moreover, the fact that the underlying paradigm is imperative and focuses on state manipulation with a lot of in-wiring of dependencies, makes it difficult to change the programs once written and to distribute the computing power over different, independent, parallel nodes or devices. The field of declarative programming with the two components of functional programming and logic programming offers a much higher level of abstraction than traditional imperative languages. The abstraction mechanisms and the adoption of functions as first-class values allow for the easy development of Domain Specific Languages (DSL) in functional programming languages. Using a DSL enables experts in the field to program in a very concrete programming language tailored for the specific domain without requiring them to learn programming in full depth. Moreover, functional reactive programming has the promise of being very suitable for handling large and heterogeneous streams of events. Logic programming languages on the other hand allow data to be represented both extensionally and intentionally and have the advantages that they are well suited for rapid interpretation of the data structure and the code to implement very complicated ideas. Due to their compact syntax and logical nature experienced programmers find the code simple to read and debug. The following challenges in robotics programming remain: - Reduce the amount of low-level robot programming - Make it approachable to non-experts - Automate the deployment of robot control software (code generation) - Links with AI - Allow the robot to automatically reconfigure itself (see above for code generation) To solve those challenges the sub-parts need to be thoroughly evaluated, expanded and aligned with the current state of the art; and finally integrated into an overarching framework. As the research is embedded in the research group DTAI, we only focus on the declarative parts." "Declarative experimentation for machine learning." "Hendrik Blockeel" "Informatics Section" "With this project we introduce the concept of ""declarative experimentation"". The motivation for declarative experimentation is that empirical testing of a hypothesis currently demands a lot from the scientist, in terms of setting up experiments, analyzing the results, and ensuring all this is done correctly. This process is not only labor-intensive but also error-prone. These problems would be solved if a scientist could formulate a hypothesis as a query in some formal language, after which the remainder of the experimental process would be performed automatically. This is somewhat similar to how people query databases by formulating questions declaratively, instead of programming the retrieval procedure themselves; hence the name ""declarative experimentation"". While it may not be realistic to assume that declarative experimentation is possible in all sciences, we are convinced it is possible and desirable in the area of machine learning, and likely also in many other experimental sciences, including bioinformatics. In this project, building on the already existing concept of experiment databases, we will develop a theory and methods for declarative experimentation in machine learning. The results may serve as a starting point for developing declarative experimentation in other areas." "Studies on declarative process modeling and its relation to procedural techniques." "Jan Vanthienen" "Information Systems Engineering Research Group (LIRIS) (main work address Leuven)" "The challenging task of managing business processes has become an even more complex endeavor as companies are required to sustain flexible and agile business practices. Process management research has proposed numerous ways of accommodating for this need for flexibility. The declarative process modeling paradigm, with its shifts towards a constraint-based way of approaching process behavior, is a prime example. The downside of numerous languages covered by this paradigm, however, is the complex nature of their constraints, as well as their interactions. This thesis offers numerous solutions to overcome these impediments for improved usability of declarative process models.In the first part, an overview is given of the current landscape of declarative process modeling, extended with the formal basis for the remainder of the text. A conversion of a common body of process constraints into a procedural variant is provided as well.In the second part, an approach to reveal (hidden) dependencies among constraints in declarative process models is proposed. This contribution allows users to better grasp the full behavior of such models, for implicit connections are explicited and added as an extra layer of annotation on top of the current representation. The effectiveness and usefulness of the approach is illustrated in a user study and is also used for constructing a complexity measure for constraint-based models.In the third part, the comparison with the procedural process modeling paradigm is made. Analogies and intricacies to both approaches are leveraged towards modeling in a mixed-paradigm fashion, as well as towards achieving better automated process discovery results. The findings are further extended by an approach for checking mixed-paradigm models for inconsistencies, and a conformance checking approach for assessing mixed-paradigm mining results.Finally, the last part provides an outlook for future work.All examples are elaborated in the Declare framework, which provides a widely-supported language and body of tools." "A declarative framework for clustering." "Hendrik Blockeel" "Informatics Section, Declarative Languages and Artificial Intelligence (DTAI)" "One of the key tools to gain knowledge from data is clustering: identifying groups of instances that are highly similar. In the traditional unsupervised clustering process, a practitioner needs to select a suitable distance measure, clustering algorithm, and corresponding hyperparameter settings to arrive at a satisfactory clustering. Making the right choices for these components is hard, and as a result a practitioner often performs many iterations in search of a clustering that matches their interests. The goal of this thesis is to develop methods that make it easier to find a good clustering. Most of our contributions are in the setting of semi-supervised clustering. Semi-supervised methods allow the user to guide the clustering process through direct feedback, rather than by tedious tweaking of the components of the clustering pipeline. We consider feedback in the form of pairwise constraints, which specify whether two instances should be in the same cluster or not. We present five main contributions. We first investigate the use of internal quality measures for unsupervised algorithm and hyperparameter selection in clustering. We identify several important limitations of these measures, and conclude that they are not suited for this purpose. In our second contribution, we examine the use of constraint set consistency, a measure that captures the utility of constraint sets, for selecting a semi-supervised algorithm. Our results show that constraint set consistency cannot be used within the scope of individual datasets, and consequently cannot be used for algorithm selection. In our third contribution, we use constraints to select and tune an unsupervised clustering algorithm. We find our simple strategy to be highly effective, outperforming existing semi-supervised clustering methods that operate within the bounds of a single algorithm. In our fourth contribution, we introduce the concept of super-instances, and two concrete methods that exploit them: COBRA and COBRAS. The latter is the first semi-supervised clustering method that allows for truly interactive and iterative clustering with pairwise constraints. Finally, in our fifth contribution, we show that COBRAS can easily be adapted to take label queries into account and that this can result in better clusterings, especially when the number of clusters is large." "Declarative methods in computer science" "Applied Computer Science Lab, Databases and Theoretical Computer Science" "This Network aims to bring together researchers involved in computer science research that is either fundamental; orapplication-oriented but based on fundamental research; or both, using declarative, i.e., logic-based methods. The spectrum of researchtopics covered by this definition is quite broad, ranging from functional, logic, and constraint programming over various aspects ofartificial intelligence, including machine learning and knowledge representation and reasoning, to data mining, and various aspects ofdata science, including database research in the widest sense of the word (aiming at novel platforms and applications, such as the web).The communal theme, namely a foundational approach using logic-based methods, is the strongly unifying element of this Network" "DECIDER PRO: Decision-Centric Declarative Process Development" "Estefanía Serral Asensio" "Information Systems Engineering Research Group (LIRIS) (main work address Brussels), Information Systems Engineering Research Group (LIRIS) (main work address Leuven)" " Unlike procedural processes, knowledge-intensive processes are less structured, and dependent on the decisions and the actions of knowledge workers. Typical knowledge-intensive processes are healthcare processes and clinical pathways. Capturing these processes in a procedural and structured process model fails to provide the required flexibility inherent to the decisions made and the actions performed by the knowledge workers. Declarative process notations, most notably the language CMMN, do provide a degree of flexibility: process elements are loosely coupled by constraints that define what kind of behavior can, must, and must not happen, rather than prescribing the path that must be followed, as is the case in procedural processes. For modeling business decisions, the Decision Model and Notation (DMN) standard has proven to be suitable both in academia and in industry. However, to capture both the decisiondriven and the declarative nature of knowledge-intensive processes, these two notations need to be consistently integrated. This project will integrate CMMN and DMN for the modelling, the distributed execution of the integrated model, and the mining of such decisioncentric declarative processes. The project includes international collaborations with leading researchers in the field and will apply the methods that will be developed on healthcare processes provided by the American College of Obstetricians and Gynecologists." "Towards Declarative Statistical Inference" "Hendrik Blockeel" "Informatics Section" "Wide-ranging digitalization has made it possible to capture increasingly larger amounts of data. In order to transform this raw data into meaningful insights, data analytics and statistical inference techniques are essential. However, while it is expected that a researcher is an expert in their own field, it is not self-evident that they are also proficient in statistics. In fact, it is known that statistical inference is a labor-intensive and error-prone task. This dissertation aims to understand current statistical inference practices for the experimental evaluation of machine learning algorithms, and proposes improvements where possible. It takes a small step forward towards the goal of automating the data analysis component of empirical research, making the process more robust in terms of correct execution and interpretation of the results.Our first contribution is a synthesis of existing knowledge about error estimation of supervised learning algorithms with cross-validation. We highlight the distinction between model and learner error, and investigate the effect of repeating cross-validation on the quality of the error estimate.Next, we focus on the evaluation of multi-instance learning algorithms. Here, instances are not labeled individually, but instead are grouped together in bags and only the bag label is known. Our second contribution is an investigation of the extent to which conclusions about bag-level performance can be generalized to the instance-level. Our third contribution is a meta-learning experiment in which we predict the most suitable multi-instance learner for a given problem.The intricate nature of statistical inference begs the question whether this aspect of research cannot be automated. One requirement for this is the availability of a model representing all relevant characteristics of the population under study. Bayesian networks are a candidate for this, as they concisely describe the joint probability distribution of a set of random variables, and come with a plethora of efficient inference methods. Our last contribution is a theoretical proposal of a greedy-hill climbing structure learning algorithm for Bayesian networks." "Co-evolution of Knowledge Bases: Languages and Tools" "Joost Vennekens" "Declarative Languages and Artificial Intelligence (DTAI)" "Useful domain knowledge resides in the heads of key employees, but it remains to a large extent tacit. Storing this knowledge explicitly in a so-called knowledge base (KB) has many advantages. However, the process of building a KB by having a knowledge engineer repeatedly interview the domain experts, is slow and costly. To alleviate this, we envision a co-evolution process where the domain expert and knowledge engineer co-create the KB together, with the additional advantage that the domain expert “retains ownership” of the KB. To facilitate co-evolution, a formal language that is both understandable for domain experts and expressive enough to represent complex knowledge is needed. In this project, we will investigate 3 potential formalisms to facilitate co-evolution: cDMN (an extension to DMN), abstract syntax trees and controlled natural language. After the KB is co-created, it should be maintained, preferably by the domain experts themselves. For this task, we will investigate verification techniques for the proposed formalisms, test case generation techniques to validate the KB, and the use of inductive logic programming to automatize refinement of the knowledge. This project will develop and evaluate the proposed techniques in the context of five case studies currently executed by the research group, covering a variety of domains. If successful, this will enable domain experts to co-create the KB, perform the maintenance task, and reclaim ownership of their knowledge."