Anil Akbay Mehmet

Developing Efficient Routing Algorithms for Sustainable City Logistics

Increasing environmental concerns and legal regulations have led to the development of sustainable technologies and systems in logistics, as in many fields. The adoption of multi-echelon distribution networks and environmentally-friendly vehicles in freight distribution have become major concepts for reducing the negative impact of urban transportation activities. In this line, this thesis addresses a two-echelon electric vehicle routing problem (2E-EVRP) as a practice of sustainable city logistics. In the first echelon of the distribution network, products are transported from central warehouses to satellites located in the surroundings of cities. This is achieved by means of large conventional trucks. Subsequently, relatively smaller-sized electric vehicles distribute these products from the satellites to demand points/customers in the cities as they are less noisy and have zero direct emission. The problem addressed in this study takes into account the limited driving range of electric vehicles that need to be recharged in charging stations when necessary. Furthermore, we considered realistic charging scenarios such as partial recharging and non-linear battery charging time and observed their effect on the solution quality. In addition, the proposed problem considers some of the essential real-life constraints, i.e., time window and simultaneous pickup and delivery. A mixed-integer linear programming formulation is developed, and small-sized instances are solved using CPLEX. Due to the complexity of the problem, we have developed solution algorithms based on variable neighborhood search and construct, merge, solve and adapt to solve large-sized problem instances.

Georgios Athanasiou

Machine Learning Platform for Assisted Reproductive Technologies

In-vitro fertilization (IVF) is a domain that faces large problems in efficiency and efficacy, for human patients and mammals. So far, even though there are critical advancements in the applications of Artificial Intelligence (AI) and Machine Learning (ML) in Healthcare, but also more specifically in assistive reproduction, the use of these kinds of techniques remains low. The reasons lie in different factors, from the unsatisfactory accuracy rates to the lack of mathematical evidence for the results of some applications, and the mistrust of correctness by Healthcare scientists and professionals. The objectives of this research are the design and development of software, based on AI and ML techniques, to further support the research in assisted reproductive technologies (ART) and answer some of the existing doubts on the current applications. The contribution of this project involves (i) the design of new machine learning algorithms, (ii) the facilitation of the acquisition, analysis and validation of new knowledge through the analysis of ART data, and (iii) the introduction of new, proven AI methods for the evaluation of IVF.

Davide Audrito

Modeling the legal institutions of human and fundamental rights following a formal ontological approach

Over the last four decades, formal ontologies have represented well-rooted and effective knowledge modeling and management frameworks, which encounter the growing need to digitalize legal sources and promote interoperability. For this purpose, manifold research efforts have been put into the representation of legislation, policies, case law, administrative procedures, and further sources of legal nature, which cover manifold domains, including privacy, tenders and procurements, licenses, policies, and interdisciplinary areas. The will to pursue this research line originates from the almost complete absence, in the scientific literature, of ontologies engaged with the representation of human and fundamental rights, whose evolving nature in the digital era requires increasing in-depth analysis. Following a careful survey of the state of the art and the recognition of the most recent trends, this research effort aims to improve the semantic search for information retrieval by modeling the legal institutions concerned with the protection of human and fundamental rights. For this purpose, the dataset will include sources of conventional and constitutional nature by following a multilevel and multilingual approach. Starting from broader and generic categories of concepts, e.g. “President” or “Court”, the ontology will foster the understandability of human and fundamental rights oriented to manifold categories of users, and set an extensible framework at the crossroads of legal harmonization and interoperability.

Dimitra Anna Bourou Bourou

A formal model of sense-making using image schemas and conceptual blending

Sense-making is a process seldom adressed in AI, while cognitive science approaches it as the process of an autonomous agent bringing its own original meaning upon its environment, and proposes it is fundamental for cognition. We therefore model the sense-making process as the conceptual blending of image schemas with a structural description of a stimulus. The case study we have used is diagrams and their geometric configurations. Image schemas comprise mental structures abstracting the invariances of repeated sensorimotor contingencies such as SUPPORT, VERTICALITY and BALANCE. They structure our perception and reasoning by transferring their structure to our percepts according to the principles of conceptual blending. In our work we model the conceptual blend of various image schemas with the geometry of a diagram, obtaining a blend that reflects the interpreted diagram. We formalize image schemas and geometric configurations with typed FOL theories, and, for the latter, use Qualitative Spatial Reasoning formalisms to describe the topology, shape, and relative position of its components. Conceptual blends are computed as category-theoretic colimits. The resulting blend has emergent structure, representing a meaningful diagram, e.g. The Hasse diagram (representing a poset) as a SCALE with levels, minimum and maximum elements etc. Our work on diagrams can provide guidelines for effective visualizations, and our general framework can be developed into a system that constructs possible conceptual meanings for various stimuli types.

Camilo José Chacón Sartori

Improving Optimization Algorithms with Deep Learning and Graphical Tools

Our research work is composed of two interconnected research areas.
First, a recent line of research in optimization, which explores the potential of integrating deep learning techniques with optimization algorithms, is gaining ground in combinatorial optimization. Our objective is to employ deep learning to acquire the ability to generate practical solutions for a general problem, rather than specializing in generating good solutions for a specific case of a particular problem.
Second, another topic that will be studied in this thesis is improving our understanding of the behaviour of optimization algorithms such as metaheuristics. In general, there is a lack of accessible tools to analyse, contrast and visualise the behaviour of metaheuristics when solving optimization problems. To help to answer these questions, a recent tool called search trajectory networks (STNs) was presented in the related literature. STNs is a data-driven, graph-based tool to analyse, visualise and directly contrast the behaviour of different types of metaheuristics. However, the usability of this tool is still very much limited. One of the goals of this thesis is to improve the initial STNs tool and its usability.

Núria Correa Mañas

Mindful AI for ART: Integrating expert knowledge to trainclinically coherent models for dose selection in IVF processes

Much like a lot of fields in healthcare, Assisted Reproduction generates lots of data that up until now has gone largely unused. Even after approximately 40 years of the first successful In Vitro Fertilization treatment and all the gigantic strides made to advance Assisted Reproduction Techniques until today, the chances of achieving pregnancy after In Vitro Fertilization remain around 30%. With the aim to improve that ratio, this research will focus on applying Artificial Intelligence and Machine Learning techniques to unfold the power of the high quantity of data generated by Assisted Reproduction Techniques. Specifically designing a modular Decision Support System aiming at improving efficacy after Controlled Ovarian Stimulation, and assessing pregnancy probabilities and multiple pregnancy risk after embryo transfer.

Thiago Freitas Dos Santos

A Multi-scenario Approach to Continuously Learn and Understand Changes in Norm Violations

Using norms to guide and coordinate interactions has gained tremendous attention in the multiagent community. However, new challenges arise as the interest moves towards dynamic socio-technical systems, where human and software agents interact, and interactions are required to adapt to changing human needs. For instance, different agents (human or software) might not have the same understanding of what it means to violate a norm (e.g., what characterizes hate speech), or their understanding of a norm might change over time (e.g., what constitutes an acceptable response time). The challenge is to address these issues by learning to detect norm violations from the limited interaction data and to explain the reasons for such violations. To do that, we propose a framework that combines Machine Learning (ML) models and incremental learning techniques. Our proposal is equipped to solve tasks in both tabular and text classification scenarios. Incremental learning is used to continuously update the base ML models as interactions unfold, ensemble learning is used to handle the imbalance class distribution of the interaction stream, a transformer-based model is used to learn from text sentences, and Integrated Gradients (IG) and LIME are the interpretability algorithms. We evaluate the proposed approach in the use case of Wikipedia article edits, where interactions revolve around editing articles, and the norm in question is prohibiting vandalism. Results show that the proposed framework can learn to detect norm violation in a setting with data imbalance and concept drift.

Athina Georgara

Trustworthy Task Allocation for Human Team

In many practical applications, we notice a shift towards teamwork and collaboration. An individual alone may not have the complete skill set or the power to fulfil the requirements of a job on time. Instead, employing a group of people to join forces and carry out a given job may result in positive outcomes. However, forming efficient teams, i.e., putting together the right people that can efficiently work with each other, combine their skills, knowledge and expertise and fulfil the job assigned to them at the best possible quality, is a challenging task. A team's efficiency depends on various factors. First and foremost, it depends on the job or the task the team needs to carry out. It also depends on each team member's area and level of expertise, personality, personal preferences, motives and aspirations, interpersonal relations with the rest of the team, etc.
This dissertation addresses the problem of forming human teams, and we contribute to the problem of allocating tasks for human teams by developing tools to aid the process.
First, we explore the several components that affect the behaviour and performance of individuals when they participate in a team. To do so,  we thoroughly review the literature regarding teamwork and collaboration, considering research in the fields of Psychology and Social Sciences. Then, we formally model how to assess the impact of individuals' characteristics on a team's collective behaviour; and therefore determine the team's expected performance when they work on a specific job or task.
Second, we define the non-overlapping many teams to many tasks allocation problem (NOMTMT-AP). This is the problem of forming non-overlapping teams so that each team is allocated to work on one task while each task is tackled by one team. Third, we develop a linear programming encoding for optimally solving the problem. However, as the number of agents and tasks grows, solving the problem optimally with the means of a linear program becomes inefficient. Thus, in addition, we develop an anytime heuristic algorithm, called Edu2Com, that yields high-quality solutions to the NOMTMT-AP.
Moreover, in an era where many hard and complex procedures are automated with the aid of artificial intelligence, the need for humans to understand the rationale behind AI decisions becomes imperative. Given an artificially intelligent tool for forming human teams, explaining why and how the tool is making certain decisions is essential. Adequate explanations for decisions made by an intelligent system do not just help describe how the system works; they also earn users' trust. Towards trustworthy team formation tools, in this dissertation, we also propose a general methodology for justifying why a team formation algorithm at hand formed certain teams and disregarded others. We devise an algorithm that wraps up any existing team formation algorithm and builds justifications regarding the teams formed without modifying the team formation algorithm. Our algorithm offers users a collection of commonly-asked questions within a team formation scenario and builds justifications as contrastive explanations. In addition, we investigate privacy issues upon providing explanations in the context of team formation.
These tools can be used in a good deal of real-life applications. For instance, in this work, we have employed our tools to aid teachers from student teams to undertake a school/university project, form teams in crowdsourcing events, and form teams of volunteers to carry out social impact tasks.

David Gomez Guillen

Improving simulation model calibration for Cost-Effectiveness Analysis via Bayesian methods

The use of mathematical simulation models of diseases in economic evaluation is an essential and common tool in medicine aimed at guiding decision-making in healthcare. Cost-effectiveness analyses are a type of economic evaluation that assess the balance between health benefits and the economic sustainability of different health interventions. One critical aspect of these models is the accurate representation of the disease's natural history, which requires a set of parameters such as probabilities and disease burden rates. While these parameters can be obtained from scientific literature, they often need calibration to fit the model's expected outcomes. However, the calibration process can be computationally expensive and traditional optimization methods can be time-consuming due to relatively simple heuristics that may not even guarantee feasible solutions. This thesis investigates the use of Bayesian optimization to enhance the calibration process by leveraging domain-specific knowledge and exploiting inherent structural properties in the solution space.

Andrea Guillén

The human-centric approach in Industry 5.0: legal and ethical challenges

While Industry 4.0, is focused on technological development, Industry 5.0 embraces the societal and environmental factors of the future of work at factories. This transition from digitalization and automation to Industry 5.0 relies on three aspects: human-centricity, sustainability, and resilience.
The human-centric approach of Industry 5.0 aims at promoting the wellbeing of industry workers by placing workers needs and interests at the core of the production processes and preserving their fundamental rights. This raises the question of how to achieve human-centricity in smart factories, where digitalization and automation have introduced new legal and ethical concerns, and exacerbated existing ones. In particular, due to the deployment of data-driven technologies, including Artificial Intelligence, Internet of Things, Digital Twins, Augmented Reality and Blockchain in the workplace.

Bjoern Komander

Automated Monitoring of Online Communities

Online communities have become a vital platform for users to express and discuss a wide range of opinions and topics. However, as these communities grow in size and content, they face challenges related to information overload, misinformation, hate speech, and polarization. Recent advances in Natural Language Processing (NLP) and Graph Neural Networks (GNN) offer promising solutions to automatically analyse, monitor and enhance online communities. This PhD project aims to develop tools for the automated analysis and enhancement of online communities by leveraging the relational information within these communities and the text data. The objectives include developing novel methods for analyzing online communities, improving our understanding of these communities, integrating automated tools to facilitate and monitor discussions, and assessing the usefulness of these tools. An important focus is to develop novel experiments that can evaluate the impact of newer technologies, such as tools based on large language models.

Roger Xavier Lera Leri

Explainability for optimisation-based decision support systems

In the last years there has been an increasing interest in developing Artificial Intelligence (AI) systems centered in humans that are trustworthy, meaning that they have to be ethical, lawful and robust. Within this new vision of AI, there is an strong consensus to require explainability in AI algorithms, i.e. the capacity to provide explanations of the decisions taken by such algorithms.
Hence, the goal of this PhD thesis is to develop decision support systems that not only recommend the optimal solutions for different real-world problems, but also to develop a general framework for explainable AI that provides explanations of the decisions taken by our approaches. We aim at formalising such problems as convex optimisation problems, enabling the use of commercial-off-the-shelf solvers to solve such problems, using real world data instances to evaluate our approaches.

Alejandra López de Aberasturi Gómez

Multi-Agent Reinforcement Models of Human Group Productivity in Educational Settings

We use the term "Collaborative Learning" to refer to educational approaches that involve the collective intellectual effort of students in a group. Students are organized into groups of three or more members to solve a problem, find answers to a question, search for meanings of concepts, or create a product. Each group member can work on inter-connected tasks contributing to a common overall outcome. They can also work in parallel on a shared assignment.
Given a group of students and a collaborative learning task, this thesis raises the question of how to use team modeling and simulations to facilitate the performance of human groups.
We propose that a well-conceived and trained multi-agent model can be used to make predictions about the performance of a specific group on a given task and that we can leverage this predictive capability to optimize and facilitate the work of human groups.
To do this, we understand the group as a set of individuals who interact with each other on the basis of a relationship of interdependence. We start with Steiner's group productivity model and task taxonomy. Plus, we adhere to the assumption that individuals act as maximizers of expected utility. This hypothesis opens the door to using reinforcement-trained agents as a descriptive (rather than prescriptive) model of a group of individuals within an interdependent relationship, each seeking to satisfy their own selfish goals.

Nieves Montes

Value Engineering for Autonomous Agents

Value engineering consists in the formulation, design and implementation of new value-wary functionalities for autonomous agents. In this thesis, we propose perspective-dependent value-based normative reasoning, which allows agents to reason about which norms and regulations are best aligned with respect to not only their values, but to the values they estimate that others in their community have. To achieve this objective, we start from Schwartz’s Theory of Basic Human Values and establish the consequentialist nature of the norm-value relationship. Then, we contribute the Action Situation Language, a novel norm representation language rooted in institutional analysis and with deep links to game theory. Last, we introduce Theory of Mind functionalities into an existing BDI autonomous agent architecture. Together, these three contributions are integrated in a novel functionality that enables autonomous agents to reason about prescriptive norms in a perspective-dependent manner, by switching its value system to the one it estimates that another agent has at runtime. Such perspective-dependent value-based normative reasoning functionality, with its inherent social orientation, constitutes a novel contribution to the community of values for autonomous agents.

Shuxian Pan

Automated Extraction of Online Community Norms

To develop cooperative multiagent systems effectively, we aim to create an architecture that facilitates the agents' dynamic adoption of conventions. It expands an existing agent model's action selection architecture with a component that uses Natural Language Processing techniques. This component embeds conventions into agent interaction strategies to improve the predictability of other agents' actions if all agents adopt the same conventions in their strategies.

Elham Ali Rababa Rababah

Enhancing branch and bound with conflict-driven clause learning

The Boolean Satisfiability problem (SAT) is the problem of deciding whether a Boolean propositional formula can be satisfied. One of the most popular applications of SAT is its use as a logic-based formalism to express problems of industrial interest as a set of constraints that must be satisfied. The success of SAT as a problem-solving formalism is notorious due to the high efficiency of current SAT solvers, which let to obtain high quality solutions with competitive computation times, in spite of the extreme hardness of the tackled problems. The efficiency of SAT solvers can be mainly attributed to the use of clause learning techniques. Recent works have shown that we can make further use of clause learning by combining it with Branch and Bound in the particular case of the MaxSAT problem. In this thesis, we aim at improving and extending the use of Branch and Bound to the wider range of combinatorial problems. We start from the observation that Branch and Bound with clause learning is applicable to any problem that can be stated in terms of pseudo-Boolean (PB) constraints -a particular kind of arithmetic constraints. Therefore, the challenge that we face in this thesis is to obtain a practical benefit from this technique in different problems containing PB constraints, by developing efficient algorithms to be integrated into SAT solvers. We will identify scenarios where we can improve the state of the art in terms of required solving time thanks to the new approach. As first steps, we consider the family of scheduling problems, where PB constraints with favorable properties abound.

Guillem Rodriguez Corominas

Algorithms and techniques for computer assisted anastylosis

The term anastylosis refers to the reconstruction of antique artifacts---such as sculptures, frescoes and wall paintings---from their fragments. Anastylosis was an activity that was exclusively performed manually, by experienced personnel such as archeologists. However, in the presence of a large number of fragments, manual anastylosis is a very tedious and time-consuming activity which is often made more difficult by eroded fragments, missing fragments, and fragments that have lost their original color. Moreover, several authors have mentioned the problem of additional fragment erosion due to repeated handling for the purpose of manual anastylosis. Therefore, the computer science and artificial intelligence communities have started to develop computer-assisted methods for anastylosis.
Computerized reconstruction generally involves two main tasks: local assembly and global assembly. Local assembly focuses on identifying neighboring fragment pairs and determining their transformations, while global assembly utilizes the results obtained from local assembly to produce globally optimal fragment placements.
The primary objective of this thesis is to improve the existing algorithms and techniques used for computer-assisted anastylosis, addressing the inherent challenges.

Gerard Rovira Parra

Machine Learning Support for in-class Competence-based Learning

The main objective of the model presented here is to monitor the learning progress of each student, especially in the face-to-face context, guiding the teacher in the often described busy pace required by the practices of formative assessment. This is achieved by providing the necessary feedback to each student and by concentrating on-site observation on those students who need more support, through the paradigm known as the "teacher-as-a-sensor" (TaaS), the idea of which is to use the teacher's own specialised observation criteria to avoid introducing elements that violate the privacy of the students in the classroom.
The model has some of its roots in cognitive science based on neuroscience in the Atkinson-Shiffrin Model of Memory and D. Willingham's diagram of the mind, which has led to a division of learning into four fundamental pillars: behaviour (including motivation and satisfaction), attention, thinking/comprehension and knowledge. The machine learning models that drive the overall model consist of simple algorithms tuned to the small amount of data expected. The machine learning is enhanced with online machine learning techniques and ensembles that combine the different learned evaluation criteria of each teacher, obtaining a real-time analysis of the learning status of each student and adapting to the main target of the model.
In the context of data mining, five main foundations have been identified for the evaluation of the four pillars representing learning, which partially consist of specialised activities (clicker questions, concept maps, easter eggs, etc.), that are proposed by the system depending on data availability, and accompany the existing ones in the Additio App library and can be extended in the future from the online platform.

Borja Velasco Regulez

Causal analysis algorithms in healthcare-related settings

Many studies in the healthcare field and in the Health Technology Assessment (HTA) field are aimed at answering causal questions. Causal questions are those that revolve around causes and effects. Example: Does putting antibiotic in the knee during knee arthroplasty reduce the risk of prosthetic infection? Does the vaccine against COVID-19 cause alterations in the menstrual cycle? These are causal questions. These type of questions are often addressed by healthcare and Health Technology Assessment (HTA) researchers with causality-free approaches, based on associational risks or correlations. This is an epistemological limitation, and it has been proven that it can lead to biased conclusions. The most suited approach for answering these type of questions is causal analysis.
We aim to develop, compare and apply causal analysis algorithms in healthcare related settings, especially in HTA, using both synthetic and real-world datasets. From a methodological perspective, we focus on the use of neural networks and other machine learning algorithms for doing causal analysis, as well as on the topic of multivalued treatment settings, which is traditionally neglected in the literature. From an applied perspective, we employ real-world healthcare data managed by AQuAS (Quality and Evaluation Agency of the Catalan Healthcare System) for addressing relevant questions of the healthcare field, in collaboration with clinicians and healthcare system managers and using state of the art methods with desirable properties such as doubly robust estimators.