CA | ES | EN
Doctoral Consortium 2024

The Doctoral Consortium will take place on June 27, 28 and July 1. 

Every student will have 15 minutes to explain his/her PhD progress and plans and 10 minutes more to discuss with the committee's three members. Students should bear in mind that the level of the explanation has to be understood by an audience with knowledge in AI but not in the specific area of interest and should give particular importance to the impact that the PhD can potentially have. Presentations are encouraged to be done in English unless specific exceptions.

Possible presentation format:

  1. Introduction
  2. Research Objectives
  3. State of the art
  4. Research methodology and work plan
    1. Research methodology
    2. Work plan
    3. Detailed work plan for the current academic course
  5. Publications (optional, if any)
  6. Research Stage (optional, if any)
  7. Conclusions

After the presentations, there will be a consensus meeting of the committee members to prepare the final reports.

The Schedule of the presentations is the following:

June 27




Jaume Reixach

Leveraging Machine Learning for Metaheuristics

Metaheuristics are general frameworks designed to find high-quality solutions to combinatorial optimization problems. This thesis primarily focuses on integrating machine learning techniques into existing metaheuristics, with the goal of improving their performance. Two main paradigms are explored in this context. The first paradigm involves online learning methods. So far, this has consisted of implementing a learning mechanism within the Construct, Merge, Solve and Adapt (CMSA) metaheuristic, enhancing its simplicity and performance. This implementation draws inspiration from the classic multi-armed bandit problem in Reinforcement Learning. The second paradigm enhances metaheuristics through offline learning. This approach has led to the development of an evolutionary-based framework, which has been applied for learning the heuristic function of a tree search method and improving the performance of a genetic algorithm by learning high-quality individuals which are used as guidance.


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 explores the application of Bayesian optimization to improve the calibration process, leveraging domain-specific knowledge and exploiting structural properties to efficiently handle multiple constraints in high-dimensional functions with a sequential block decomposition.




Roger Xavier Lera Leri

Explainability for optimisation-based decision support systems

In recent years, there has been an increasing interest in developing Artificial Intelligence (AI) systems centred on humans that are trustworthy, meaning that they have to be ethical, lawful, and robust. Within this new vision of AI, there is a 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, and using real-world data instances to evaluate our approaches.


Alejandra López de Aberasturi Gómez

Reinforcement Learning Models of Human Teamwork Productivity

Humans possess innate collaborative capacities. However, effective teamwork often remains challenging. Social psychology and game theory provide valuable insights into the factors influencing voluntary collaboration in teamwork settings. This thesis delves into the feasibility of collaboration within teams of self-interested agents who engage in teamwork without the obligation to contribute. By integrating insights from aggregative game theory, reinforcement learning and social psychology, we seek to model realistic teamwork settings and to design a framework that learns approximations of realistic teamwork strategies. This research can contribute to the development of evidence-based approaches for designing teamwork settings that favour cooperation in teams and societies of humans and/or machines where cooperation is not enforced


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 applies Natural Language Processing techniques with a domain-specific knowledge base. This component embeds conventions into agent interaction strategies to improve the predictability of other agents' actions. At the same time, Natural Language Processing techniques allow the users to introduce the conventions or provide their domain-specific knowledge to the multiagent system in a more user-friendly way.




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 examples of health-related 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 a 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/inference. We aim to develop 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.


Jairo Alejandro Lefebre Lobaina 

Explainable argumentation based multi-agent recommender system

In recent years, there has been a growing emphasis on the development of Artificial Intelligence (AI) systems that prioritize human-centric design, emphasizing trustworthiness, ethics, legality, and robustness. Central to this evolving paradigm is the demand for explainability within AI algorithms, whereby these systems can articulate the rationale behind their decisions. The primary objective of this research is to design and implement a multi-agent recommender system that allows to combine the recommendations and associated explanations provided by the agents. Our aim is to formalize how to explain the final recommendation by using all recommendations provided by several agents within the context of argumentation theory. Given a set of recommendations and associated explanations, the expected output is a set of sorted recommendations base on their relevance. The sorted recommendations will also be supported by explanations associated to the reasoning process, while employing authentic datasets to validate our approaches.


Arnau Mayoral Macau

Design of approximately ethical environments

Reinforcement learning (RL) algorithms have seen many applications, from game playing to autonomous driving and conversational agents, demonstrating a remarkable ability to learn complex tasks. However, while these algorithms may excel in learning to optimize a given reward function, ensuring that they behave according to human moral values while doing so is a significant challenge. On my thesis, we are working on an algorithm to automatically design ethical RL environments with large state spaces in such a way that the optimal behavior for the agents to learn is value aligned. We do so by adding an extra objective, called ethical objective, with its own system of rewards that incentives ethical actions and punishes unethical behavior. Then, our algorithm computes a so called ethicalweight which will be used to linearly combine the individual task of the environment with the ethical objective. With this linear combination we transform a multi-objective problem into a single-objective problem, where agents learn with just one reward function. Our contribution is how to compute the minimal weight needed to make the ethical objective prevail over the individual objective, so when agents learn using the combined reward function, all agents learn to do their task while abiding by the moral values encoded in the ethical objective.


Gonçalo Melo de Magalhães

Smart Design Teams: Enhancing High-Performance Design Teams with AI

This PhD project investigates the optimization of team formation in digital product design using metaheuristics. It addresses challenges in team dynamics, including remote collaboration and integration of new technologies. The research leverages metaheuristic techniques to optimize team compositions, enhancing efficiency and innovation. A systematic literature review and empirical validation form the methodology, focusing on AI's role in recruitment and team assembly. The goal is to understand what has been done in terms of metaheuristic application in similar areas, and develop a framework to improve decision-making and team dynamics, potentially revolutionizing digital product design. Key areas of focus include understanding current knowledge, applying metaheuristic analysis, and empirical testing with digital design teams. The project's findings could offer significant contributions to both academic research and practical applications in digital product design.


June 28


Camilo José Chacón Sartori

Enhancing Optimization Algorithms with Learning Techniques and Graphical Tools 

Our research addresses two interconnected areas: (1) Enhancing Metaheuristic Solutions with Learning Methods: We focus on improving the quality of metaheuristic solutions by incorporating advanced learning techniques such as Graph Neural Networks. Recently, we've started leveraging Large Language Models (LLMs) as pattern detection tools to assist researchers in identifying valuable information for refining metaheuristics. (2) Understanding Metaheuristic Behavior with Visual Tools: Given the inherently stochastic nature of metaheuristics, understanding their behavior across different problems is crucial. Visual tools play a key role in enriching numerical analysis. My research aims to enhance Search Trajectory Networks (STNs), a tool introduced in 2021, by adding new visualization mechanisms and making it accessible via its web version, STNWeb. Additionally, we are integrating LLMs to generate automatic reports of these graphics, which can be complex to interpret without specialized knowledge.


Daniel Pardo Navarro

AI-Powered cheese manufacturing: a digital future for dairy

Milk is a very important food worldwide due to its high nutritional value and is present in the regular diet of most households. As a raw material it has various possibilities of transformation into products that retain its nutritional characteristics such as yogurt, cheese, butter, powdered milk, milk cream, and sour cream, among others, that allow their useful life to be increased. Its composition may present variability due to various factors such as diet, lactation stage, genetic parameters, health and well-being of the animals. This variability determines the final use of the milk and the products obtained, since it affects quality and performance. Among the derived products, cheese is the most important due to its direct consumption and as an ingredient in processed foods. Its production is based on different biochemical processes, where coagulation, syneresis or ripening stand out as critical stages.




Stephanie Malvicini

Characterization and quantification of misinformation in social platforms content

Misinformation is becoming a big issue in current society, especially as users are exposed to information daily, on every social platform at all times. Misinformation can arise from errors or be spread intentionally and can affect the lives of people by altering their decisions and beliefs. In recent years, substantial effort has been made to identify, analyze, and prevent some of those phenomena. Still, most of them focus on one, such as fake news or polarization, on a specific area of influence, using tools that do not explode all the characteristics of the phenomena. Our work aims to create generic models to understand, quantify, and measure misinformation on social platforms. By combining state-of-the-art natural language processing, machine learning, argumentation, and knowledge-based models, among others, we seek to provide a holistic approach. Additionally, we aim to create tools to simulate the flow of misinformation and user interactions in the context of social platforms.


Elifnaz Yangin

Inference systems for MaxSAT

The Boolean Satisfiability problem (SAT) is the paradigmatic NP-complete problem. This is the problem of deciding whether a Boolean propositional for- mula can be satisfied. MaxSAT, is an optimization version of SAT that aims to find an assignment that maximizes the number of satisfied formulas of a given multiset of propositional formulas. These problems are significant because many practical problems can be reduced to them and solved using an off-the-shelf SAT or MaxSAT solver. While SAT is employed to solve decision problems, MaxSAT is utilized to solve optimization problems. Despite the potential of MaxSAT resolution in solving combinatorial optimization prob- lems, it has not yet been thoroughly explored from a practical perspective. Our aim is to fill this gap, as well as to study new inference systems for MaxSAT and MinSAT, where MinSAT is the dual problem of MaxSAT and its goal is to find the minimum number of clauses that can be satisfied in a given multiset.


Rocco Ballester

Harnessing Quantum Computing for Advancements in Federated Learning and Optimization

The rapid advancements in quantum computing are opening up new possibilities in areas like machine learning and optimization. Quantum computers, with their unique capabilities, can potentially revolutionize how we solve intricate computational problems. This could lead to significant improvements in the efficiency and effectiveness of machine learning algorithms and optimization techniques, enabling breakthroughs in various scientific and industrial applications. This industrial PhD program aims to investigate the integration of quantum computing techniques into federated learning, with a particular emphasis on discovering innovative methods to optimize federated learning processes. Furthermore, the research will explore the use of quantum annealers for optimization tasks, providing novel solutions to difficult optimization problems.


Ion Mikel Liberal

Proof complexity beyond Resolution

The Boolean satisfaction problem (SAT) consist on determine if a propositional formula can be satisfied in the context of classical logic. It has purely logical motivations, but it acquires significant relevance in computer science. In practice, the solving of this problem has particular interest in industry since many industrial processes can be modelized and studied as SAT instances. Many of the research in solving SAT has been done in terms of practical approaches, however, the theoretical study of proof systems (which are theoretical frameworks to conclude the satisfiability of such formulas) has also many applications on the resolution of this problem. In proof complexity, as we have mentioned, one studies the complexity of the proofs in different proof systems. Usually, the main motivation in this area is to establish a hierarchy between all the known systems in terms of the capability of polynomially simulate each other, which has a relation with the size of the proofs they give to particular formulas . Connected to this, and for example, it is well-known that the Resolution system has not efficient proofs for the Pidgeon Hole Principle (PHP), whose proofs are at least exponential in the size of the original formula. This translates to the fact that, in practice, no algorithm based on Resolution finds optimal solutions for PHP for instance. Due to this reason, studying the complexity of some other proof systems and their automatability could deliver some interesting results in the improvement of the solving of this kind of problems. Hence, the main goals of this research is to study proof systems beyond Resolution in view of practical improvements (or limitations) of SAT solving.



July 1


Bjoern Komander

Automated Monitoring of Online Communities

Online Communities are a vital platform for users to express and discuss a wide range of opinions and topics. However, these communities face challenges related to information overload, misinformation, hate speech and polarization. To materialize the positive potential of Online Communities, there is a need for automated tools that can analyse and monitor large amounts of textual and relational data. However it is not fully clear what constitutes a high quality online community, such as a high quality online discussion. In this PhD, we aim to utilize insights from deliberative theory to develop quality measures of online communities. In combination with recent advances in machine learning and Natural Language Processing we aim to develop tools to analyse and monitor Online Communities.


Alba Aguilera Montserrat

An Agent-based model for poverty and discrimination policy-making

Artificial intelligence is already a key piece in our society, helping us to foresee and support our decisions. Ensuring that intelligent systems are value-aligned and promote social good is a major concern. Simulations can provide insights into the effect of norms on society, aiding decision-makers in the design of legal policies. My research focuses on developing an agent-based model for policy-making to address SDGs through a social lens. In particular, we aim to examine poverty from the perspectives of discrimination, healthcare, education or capabilities. By integrating a set of norms, representing legal policies, into the model, we plan to observe their effects on the agents' profiles, behaviors, and environment through a set of outcomes.


Laura Rodriguez Cima

Incorporating Social Values into Automated Negotiation Strategies

In today's society, aligning AI with social values is a major concern. Researchers are exploring various approaches to ensure AI systems adhere to human values, developing technologies that not only optimise performance but also promote social well-being. Traditional automated negotiation primarily focuses on agents optimising their individual outcomes. However, introducing social values, such as fairness, complicates the negotiation process as agents must balance personal gains with collective well-being and equitable outcomes. Our research focuses on developing negotiation strategies that prioritise both individual payoffs and social values. We will explore the theory behind these strategies, propose models incorporating social values into utility functions, and present experimental results using use cases.a sets, demonstrating its ability to adapt and improve quality progressively over time.


Salvador Pulido Sanchez

Design and Evaluation of a Reinforced Learning (AR) model to improve Deliberative Quality Online

This project proposes a specific approach using neural network architectures applied to natural language processing (NLP) such as Transformers and reinforcement learning (AR) algorithms to improve deliberative quality in online discussions. The inclusion of the Deliberative Quality Index (DQI) and the design of a personalized objective function will allow obtaining a highly specific model adapted to the challenges of evaluating deliberative quality. This approach has the potential to significantly advance the understanding and application of reinforcement learning techniques in the context of digital democracy and online citizen participation. With this research work, we seek to delve deeper into the field of natural language processing (NLP), focusing on improving the deliberative quality in online discussions. A variety of NLP neural network architectures will be explored, with the aim of developing a reinforcement learning model tailored for this purpose. The aim is to define an objective and loss function that guide the model towards specific improvements in the dimensions of the Deliberative Quality Index (DQI). In addition, an experimental validation of the model will be carried out using labeled or unlabeled data sets, demonstrating its ability to adapt and improve quality progressively over time.