The Doctoral Consortium will take place on July 21 and 22. Due to concerns regarding COVID-19 the DC2020 will be held online (see the instructions at the end).
Chairs and organizers: Felip Manyà, Pedro Messeguer, Josep Puyol-Gruart, Jordi Sabater-Mir.
Committees: Eva Armengol, Filippo Bistaffa, Christian Blum, Jesus Cerquides, Lluís Godo, Tommaso Flaminio, Dave de Jonge, Jordi Levy, Felip Manyà, Pedro Messeguer, Nardine Osman, Enric Plaza, Jordi Sabater-Mir, Marco Shorlemmer, Amanda Vidal
Every student will have 15 minutes to explain his/her PhD progress and plans, and 10 minutes more to discuss with the three members of the committee. 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 special importance to the impact that the PhD can potentially have. Presentations are to be done in English.
Possible presentation squeme:
After the presentations there will be a consensus meeting of the committee members to prepare the final reports.
The schedule (provisional) of the presentations is the following:
Embodied Cognition in Making Sense of Diagrams
In this work we propose a cognitively-inspired computational model of the role of human embodied intuitions in making sense of abstract domains. To our knowledge, this aspect has not been sufficiently addressed in artificial intelligence, despite its potential for improving human-computer interaction technologies. We hereby propose to account for this embodied knowledge using image schemas and examine the case of diagrams for ontology engineering. Image schemas are a set of primitive mental structures which generalize sensorimotor patterns a human experiences recurrently, and are transferrable to abstract domains through conceptual blending. Using the tools of algebraic specification and qualitative spatial reasoning, we can obtain mathematical descriptions of both the image schemas and the geometry of the diagram, implement the blending process, and examine whether the resulting blend is consistent with the set-theoretic semantics of the visual formalism. Such an implementation would enable making specific predictions about the intuitiveness of a diagram, which could be useful, for instance, in the context of a computational tool for science education.
Creation of a modular Decision Support System to improve results in Assisted Reproduction
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 fertility potential in sperm samples.
Fuzzy Horn clauses in artificial Intelligence: study of free models, and applications in art painting style categorization.
The main contributions of this PhD thesis are the systematic study of the universal Horn fragment of predicate fuzzy logic in the context of the general semantics introduced by Hájek (1998), and the design of an art painting style classification algorithm by using qualitative descriptors and fuzzy knowledge representation.
Combining Classical Techniques with Machine Learning for Combinatorial Optimization Problems
Combinatorial Optimization is a very important technique that has been successfully applied in many prominent scenarios (e.g., shared mobility, cooperative learning) to solve fundamental tasks such as coordination and task assignment. To tackle the complexity inherent in large-scale real-world domains, Combinatorial Optimization research has usually resorted to ad-hoc approaches that are difficult to apply to structurally different domains. The ultimate goal of this thesis is to study general approaches for Combinatorial Optimization that can be applied to different scenarios without the need of new domain knowledge. To achieve this objective, we aim at intertwining classical techniques with Machine Learning, which represents a natural candidate to surpass the above-mentioned limitation. As a prominent test-case, we are currently focusing on the Combinatorial Optimization problem inherent in large-scale online ridesharing. To attack this problem, we propose a novel approach that first generates good candidate solutions by means of Generative Adversarial Networks (GANs), and then computes the final solution by means of an Integer Linear Program (ILP).
An Ontology Based Framework for Norm Alignment
In the context of a community, norms (e.g. be helpful with the members) guide the way agents (human or artificial) will act when they interact with each other. These norms influence their expected behavior and how they will respond in a certain situation. With this in mind, in this work we are considering the use of norms in an online community by using normative systems, which are systems that regulate the interactions according to a set of formalized norms. Nevertheless, one issue that may arise in normative systems is the way different individuals interpret these norms, possibly leading to unexpected behavior in interactions, affecting the individual and the community experiences. Following these lines, the goal of this PhD thesis is to investigate the challenges associated with the failure arising in normative systems due to misalignment of the interacting agents, which is usually triggered by their diverse backgrounds and the misunderstandings they result in.
Team Formation Methods for Dynamic Large-Scale Competence Based Problems
Team formation is an optimization problem that has gained a lot of attention not only in a large variety of research fields, but also in industry. The abstract objective of this problem is to build up teams that will perform efficiently and achieve the desirable quality of the collaboration’s outcome. As the number of individuals and the number of tasks grows, the team formation problem becomes an extremely large and hard problem as it needs to consider combinations of (a) individuals to synthesize a team, and (b) teams with tasks; fact that makes finding intelligent methods essential. The problem is specialized by adopting the notion of competences and skills, and then by characterizing both the teams and tasks in such terms; with this as basic guideline, the problem becomes a competence alignment problem between teams and tasks, which allow us to device heuristics that exploit relations between competences and yield good matches. The competence-based structure can be extended and combined with several psychological and social aspects that affect the performance of a team. In this way, we are able to develop methods that have better `control' over the performance of a team, instead of depending solely on the capabilities of the team. Moving towards a broader range of problems, we see the impact of time which makes time a key parameter of the problem. Considering time can lead into two different paths: (a) scheduling the tasks and teams in a time horizon; and (b) experiencing alterations in individuals' characteristics after repeatedly engagements with different tasks and teams. As such, the main objective of this PhD thesis is to develop intelligent tools that tackles efficiently the team formation problem under a matching, scheduling, and long-term planning point of view.
Anytime Case-Based Reasoning in Large-Scale Temporal Case Bases
Case-Based Reasoning (CBR) methodology’s approach to problem solving that “similar problems have similar solutions” have proved very favorable for many industrial AI applications. However, CBR's very advantages hinder its performance as the case bases (CBs) grow more than moderate sizes. Finding similar cases is expensive. This handicap often makes CBR less appealing for today’s ubiquitous data environments while, actually, there is ever more reason to benefit from this effective methodology.
This thesis aims to speed-up the search by leveraging both problem and solution spaces in large-scale CBs where the cases are temporally related as in the example of electronic health records. We also endow CBR system with anytime algorithm capabilities to provide approximate results with confidence when the speed-up for exact search may still be not feasible. Exploiting the temporality of cases allows us to reach superior gains in execution time for CBs of millions of cases.
Automatic Analysis of Online Normative Texts
This study aims to identify and extract the subjects, verbs, objects and Conditionals implicated within the norms in normative texts to be understood towards a future formalization and translation to Logical Language. To execute this, we have proposed a mechanism based on the syntactic dependency tree in order to identify and extract the elements within the normative texts. Moreover, along the process of automatic identification and extraction, we have proposed a model based on a long list of signal words (conditional introductory words, such as, "if", "unless", and so on), and syntactic patterns that can be used to identify conditionals that represent logical implications. This is considered important through this research, bearing in mind that previous works only focus on some signal words, such as "if" and "unless", yet they do not make a complete and deep analysis of other signal words that can also introduce conditionals. As a result, a signicant amount of sentences include signal words, but do not introduce a conditional representing a logical implication. Thus, our approach, based on syntactic patterns, is mainly able to identify those cases.
Collective Intelligence in Multi-robot System
The focus of this Ph.D. research project is the mechanism and technology of auto-negotiation, which allows autonomous vehicles to interact with each other while driving on public roads. This year, our research is divided into two parts. In the first part, we propose a set of traffic road models based on the curvilinear coordinate system. This model divides the road network into multiple routes, multiple intersections, and the relationship between intersections and routes. Based on this model, we show the vehicle, the feasibility trajectory, and the conditions under which the feasibility trajectory exists, and use the theorem to prove the results. In the second part, we propose a set of multi-agent models that divide the transportation network into vehicle agents and intersection agents. Vehicle agents can choose different routes to reach the same destination, while intersection agents can choose different strategies. This strategy will have different effects on vehicles traveling at the intersection. Our research will find a balance point from the perspective of game theory so that the selection of all agents reaches equilibrium. These two parts of the research will have an important impact on the entire doctoral period, and will not lay a solid foundation for future research.
Moral values machine learning in multi-agent systems.
The main objective of this thesis is to define formally what ethical behavior is in a multi-agent system, based on both ethical theory and game theory. Specifically we want to be able to formally define what it means to be aligned with a system of moral values in multi-agent environments such as Markov Games. When formalised, we want to provide an effective algorithm for a multi-agent system to learn to behave aligned with a system of moral values.
Stochastic Natural Gradient based algorithms and convergence
Convergence property is not ensured for natural gradient based algorithms. Moreover, computational complexity is high. This usually discards these kind of algorithms when solving a particular optimization problem. The goal of this work is generalizing existing algorithm convergence results to then create new natural gradient based convergent algorithms which are computationally efficient by using Riemannian Geometry concepts, specially dually flat manifolds.
Behavioral Change by Applying Machine Learning Through Development of a Holistic Health Recommender System.
This research focuses on behavioral change to improve health habits based on a research named CarpeDiem, performed by the eHealth unit at Eurecat technology center. We consider to study four pillars of health including physical activity, sleeping, nutrition and mindfulness toward the behavioral change as the main goal of the research. Hence, research and develop a novel health recommender system would allow to achieve it.The aim is analyzing all the four dimensions of health pillars in an integrated way, and also, considering to apply Machine Learning techniques to develop a novel holistic health recommender system.
Moral Values in Norm Decision Making
Norms are well established coordination mechanisms in both agent and human societies. However, although legislators may have ethics in mind when establishing new regulations, there is a lack of explicit reference to moral values in regulatory frameworks ---which is also the case in most decision making processes. This thesis is aimed at providing both the theoretical foundations and practical mechanisms for the selection of norms so that the set of selected norms to enact within a society (the so-called, norm system) promotes the most preferred moral values in that society.
New Solving Techniques for MaxSAT and MinSAT
The Satisfiability problem, SAT for short, is the problem of deciding if there exists an assignment for the set of Boolean variables of a propositional formula that entails its evaluation as true. This historic problem is one of the most studied in Computer Science, becoming central both in its practical and theoretical aspects. It is remarkable that, according to the theory of computational complexity, SAT was the first to be proven NP-complete. On the other hand, MaxSAT is an optimization version of SAT which consists in finding an assignment that maximizes the number of satisfied clauses. Another optimization version of SAT is MinSAT, in this case the objective is to minimize the number of satisfied clauses. Both optimization problems belong to the NP-hard complexity class. The main objective of this thesis has been the advancing of the state of the art about solving computationally difficult optimization problems by reducing it to MaxSAT and MinSAT problems. To do this, we have researched about new resolution techniques, the development of more efficient solvers and the study of new encodings, as well as the introduction of new applications. The new developed resolution techniques are based on logical resolution calculus and semantic tableaux.
(date to be confirmed)
Algorithmic Framework for Making Use of Negative Learning in Ant Colony Optimization
Ant colony optimization (ACO), as most other optimization techniques based on learning, is exclusively based on learning from positive examples. Nature, however, has shown examples that indicate the usefulness of negative learning. Over the last two decades several research works have explored this topic in the context of ACO algorithms, with limited success. This thesis focuses on the development of general algorithmic framework for the incorporation of negative learning in ACO algorithms. One of the main ideas that will be explored exhaustively in this work is the employment of additional optimization algorithms for the identification of undesirable solution components. Our preliminary results have shown that the use of the ILP solver IBM ILOG CPLEX can be quite useful as negative feedback provider for an ACO algorithm for the capacitated minimum dominating set (CapMDS) problem. Moreover, we were able to improve the current state-of-the-art results on the same CapMDS problems in 10 out of 36 cases.
HOW TO CONNECT
For the Doctoral Consortium we will use the CONECTA platform, the same we have been using for the IIIA seminars.
To connect use the following address (it is the same we use for the seminars):
In order to improve everybody's experience, remember:
- If you can, try to connect using an ethernet cable directly to the router.
- Try that nobody else make an intensive use of the network during your presentation.
- IMPORTANT: When you connect, the system will ask you which quality do you want for your video streaming. Select LOW quality. By default, the system selects MEDIUM.
- If you are attending as public, turn off your webcam as soon as the presentation starts.