Doctoral Consortium 2021

The Doctoral Consortium will take place on July 20 and 21. Due to concerns regarding COVID-19 the DC2021 will be held online (see the instructions at the end).

Chairs and organizers: Josep Puyol-GruartJordi Sabater-Mir.
Committees: Eva ArmengolFilippo BistaffaChristian BlumJesus CerquidesTommaso Flaminio, Lluís GodoDave de JongeJordi LevyPedro MeseguerPablo NoriegaEnric Plaza.

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 to be done in English.

Possible presentation squeme:

  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:

July 20

9:45 Welcome
10:00

Núria Correa Mañas

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 pregnancy probabilities and multiple pregnancy risk after embryo transfer.

10:30

Manel Rodríguez Soto

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.

11:00 Coffee
11:30

Mehmet Anil Akbay

Developing Efficient Routing Algorithms for Sustainable City Logistics 

Sustainable city logistics can be defined as using environmentally friendly vehicles for freight distribution and designing multi-tier transportation structures to eliminate problems caused by freight vehicles operating in cities. In this thesis, the subject of sustainable city logistics will be discussed in the sense of the two-echelon electric vehicle routing problem (2E-EVRP), which is a variation of the well-known vehicle routing problem in the literature. In two-echelon distribution networks, products are transported from the central warehouses to the surrounding satellites by large trucks, while relatively smaller vehicles distribute goods from these satellites to the customers in the city. Electric vehicles are preferred to be used in the second echelon of the distribution network, as they are less noisy and have zero direct emission. This thesis aims to optimize 2E-EVRP considering some of the essential constraints of city logistics concepts such as time windows, satellite synchronization, partial delivery, and simultaneous pickup and delivery. Due to arising difficulties because of the multi-tier structure of the distribution network, utilizing electric vehicles with a limited driving range and constraints mentioned above, we have preferred to develop an efficient metaheuristic algorithm based on variable neighborhood search, large neighborhood search and construct, merge, solve and adapt algorithms to solve the problem.

12:00

Adrià Fenoy

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 ridesharing. To attack this problem, we propose a novel approach that first generates good candidate solutions by means of an attention based machine learning model trained in a Reinforcement Learning environment. Then, the final solution is computed by means of an Integer Linear Program (ILP).

12:30

Thiago Freitas Dos Santos

Learning for Detecting Norm Violation in Online Communities

This work focuses on normative systems for online communities. We are interested in addressing the issue that arises when different community members interpret these norms in different ways, possibly leading to unexpected behavior in interactions, usually with norm violations that affect the individual and community experiences. Thus, we propose a framework capable of a) learning the meaning of a violation (as understood by the community); b) detecting norm violations; c) and providing the violator with information about the features of their action that makes this action violate a norm. We build our framework using Machine Learning, with an ensemble approach as the classification technique. Since norm violations can be highly contextual, we train our model using data from the Wikipedia online community, namely data on Wikipedia edits. Our work is then evaluated with the Wikipedia use case where we focus on the norm that prohibits vandalism in Wikipedia edits.

13:00 Lunch
15:00

Nieves Montes

Value-guided norm synthesis in multiagent systems

In the field of multiagent systems, prescriptive norms and regulations are some of the most prevalent mechanisms to achieve conflict-free coordinated operation. As of late, however, prescriptions have also been identified as promising candidates to uphold moral values within a society of agents and ensure that their operation is ethically compliant. This thesis tackles the problem of norm design for value promotion in societies of autonomous agents. The goal is to design and implement a norm synthesis process that is value-guided (norms are evaluated for their value promotion success), agent-based (agents have an active role at crafting the norms) and socially oriented (agents negotiate with one another over the norms that will be implemented on the whole system). We advocate for a consequentialist view of norms, where the success of a set of rules at promoting some value has to be extracted by analyzing their impact on the situation being regulated. In this first part, we present a language to perform such analysis automatically. This language is complemented by an engine to build formal game theoretical models from rule configuration descriptions. Such game models can then be analyzed using standard game theoretical tools, and the expected outcomes computed and assessed in terms of their ethical compliance. Future work will focus on the norm generation process that agents need in order to make proposals for new regulations, and on the negotiation where agents make these proposals and evaluate those of others.

15:30

Athina Georgara

Team Formation Methods for Dynamic Large-Scale Competence Based Problems

Team formation is the problem of building groups of individuals that need to collaborate in order to collectively undertake a given task. The problem has been studied in several scientific areas including computer sciences, economics, psychology, and social sciences; while, at the same time, team formation within companies and organizations is usually characterised as a hard and time consuming problem for managers and human resources. This thesis tackles the combinatorial problem of (a) forming teams, and (b) matching teams to tasks, especially as the number of individuals and the number of tasks increase. Specifically we intend to devise automated methods that ease the team formation process and yield sufficiently good results in large-scale scenarios. In order to do so, we consider competencies required by the tasks that match the competencies collectively acquired by the team; while we exploit features that has been evidenced to affect teams' performance such as preferences over tasks, preferences over potential partners (that drives the social cohesion within a team), and individuals’ personality and gender.

16:00

Georgios Athanasiou

Machine Learning Platform for Assisted Reproductive Technologies

In-vitro fertilization (IVF) is a domain which faces large problems in efficiency and efficacy, for human patients and mammals. So far, eventhough 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 this kind of techniques remains low. The reasons lie on different factors, from the unsatisfactory accuracy rates to 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.

 

July 21

9:30

Dimitra Anna 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.

10:00

Jianglin Qiao

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 and intelligence facilities to interact with each other while driving on public roads, based on the graph theory, game theory, and multi-agent system. Firstly, we proposed a graph representation model that is used to formalize road network, intersection structure, and right-of-way.  Secondly, we formalize the traffic assignment problem as a game-theoretical model based on the definition of congestion game and population game.  We extend an open-source simulator, which is called AIM4, to more powerful features to simulator the autonomous vehicles driving in the road network.  Next, we will study properties of our game-theoretical model based on our simulation outcomes, such as Nash equilibrium, Social welfare, and Price of anarchy.

10:30

Borja Sánchez

Efficient and convergent natural gradient based optimization algorithms for machine learning

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, even though the convergence speed is as good as it can be if convergence is ensured. 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.

11:00 Coffee
11:30

Kian Seif

Development of a Knowledge-based Recommender System for the New Online Retail Traders 

The research focuses on investigating techniques to provide automated support and recommendations for users in a multidimensional domain toward users' behavioral change. In the first place, we focused on behavioral change to improve health habits based on a research named CarpeDiem performed by Eurecat technology center. We considered to study four pillars of health including physical activity, sleeping, nutrition and mindfulness toward the behavioral change as the main goal. Later, I consider to change the domain of research from health sector to the users' behavior in the online trading market by applying same relevant techniques such as clustering users, allocating best fit recommendations for each cluster, providing general recommendations toward more specific ones and prediction from time series data of how users change their behavior after having a recommendation to provide intelligent and automatic support for the users. The results of the research could benefit to improve automatic support for the users of platforms such as recommendation systems and automated online advisors.

12:00

Borja Velasco

Causal analysis algorithms in healthcare-related settings

Many studies in the healthcare field aim to answer causal questions. Causal questions are those that revolve around causes and effects. Example: Did the vaccine against the COVID-19 from the Astra-Zeneca company cause thrombosis in some patients? or would they have suffered thrombosis anyway, hadn't they received the vaccine?. These are causal and counterfactual questions. These type of questions are more often than not addressed by healthcare researchers with causality-free approaches, based solely on statistical 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. In this PhD thesis we aim to develop, compare and apply causal analysis algorithms in healthcare-related settings, using both synthetic and real-world datasets. Neural networks will be employed for developing or complementing  new algorithms, following one of the major trends in the field of causal analysis. Comparisons between approaches will be made, and real-world healthcare data managed by AQuAS (Quality and Evaluation Agency of the Catalan Healthcare System) will be employed, for addressing relevant questions of the healthcare field in collaboration with clinicians and healthcare system managers.

12:30

Qurat ul ain Shaheen

Argumentation-based Multiagent Recommender System

An interactive distributed model in which recommenders become active software agents that accompany the recommendations with arguments. This approach is radically diff erent with previous recommender system approaches (content-based or not) where the experiences of users are represented as passive data in a database. These recommender agents will be able then to engage in argumentative dialogues with other recommenders to agree (or not) on a recommendation to the user. A mediator may pair recommenders according to di erent criteria (e.g., similarity of pro les with the user, dissimilarity of their experiences) and then
aggregate the outcomes of the di erent dialogues into a meaningful recommendation to the user that will be duly justi ed by arguments.

13:00

Teddy Nurcahyadi

Adding Negative Learning to Ant Colony Optimization: A Comprehensive Study

Ant colony optimization is a metaheuristic that is mainly used for solving hard combinatorial optimization problems. The distinctive feature of ant colony optimization is a learning mechanism that is based on learning from positive examples. This is also the case in other learning-based metaheuristics such as evolutionary algorithms and particle swarm optimization. Examples from nature, however, indicate that negative learning—in addition to positive learning can beneficially be used for certain purposes. Several research papers have explored this topic over the last decades in the context of ant colony optimization, mostly with limited success. In this work, we study an alternative mechanism by making use of mathematical programming for the incorporation of negative learning in ant colony optimization. Moreover, we compare our proposal to some well-known existing negative learning approaches from the related literature. Our approach had been implemented to four types of subset selection problems: minimum dominating set, capacitated minimum dominating set, minimum positive influence dominating set, and multi-dimensional knapsack.  In all of the cases we are able to show that our approach significantly improves over standard ant colony optimization and over several competing negative learning mechanisms from the literature.

13:30 Closing

 

HOW TO CONNECT


For the Doctoral Consortium we will use ZOOM, as we do for the IIIA seminars.
To connect use the following address:

https://rediris.zoom.us/j/8031759956?pwd=ZWdRZmtmUlhWSTlvNWYva0dNYm1qdz09
Meeting ID: 803 175 9956
Passcode: 0k7it5

If you are presenting, please try to have the best connection that you can:
- Use an Ethernet cable direct to the router if you can.
- If you use WiFi, avoid other people making intensive use of it during your presentation.