Navigation

Heuristic design for dynamic feature selection

In some situations, information usage must be minimized, for instance in a medical diagnostic setting, where acquisitions can be invasive or dangerous for the patient. Reinforcement learning allows to optimally solve this problem, proposing at each step the modalities that will minimize the information/cost balance based on the currently acquired data not only greedily, but also accounting for future decisions. However, exploring the search space of all possible features combinations is computationally unfeasible. In this thesis, you will explore heuristics to quickly discard regions of the space that are not relevant due to their low value. In particular, you will derive confidence bounds for sampling-based heuristics that will allow to balance the exploration-exploitation trade-off.

Gabriel Bernardino

Algorithmic hiring

This is a project in the algorithmic fairness space, and is part of Horizon Europe project FINDHR. Algorithmic hiring is the usage of tools based on Artificial intelligence (AI) for finding and selecting job candidates. As other applications of AI, it is vulnerable to perpetuate discrimination. There are various subprojects that can be done here, including generation of synthetic data, study of ranking and recommendation algorithms, and external audits of actual hiring platforms.

Carlos Castillo

FINDHR project

Gender bias in online platforms

This project will analyze and study potential gender biases of online platforms. These may include job search engines, e-commerce websites, search engines, and platform economy sites for which some data could be scrapped/obtained. The specific platform is to be decided with the master student. The work will involve data scrapping/collection, measurement through fair ranking/recommendation algorithms, and analysis of potential biases found. It requires solid knowledge of Python and desire to work within a framework of data feminism, which seeks to eliminate gender biases in data collection and data processing pipelines.

Carlos Castillo

Risk-Prediction Instruments in Criminal Justice

The Justice Department of Catalonia has been using for several years structured risk prediction instruments, similar to COMPAS in the US or OASIS in the UK, to predict the risk of recidivism and violent recidivism. Correctional officers, psychologists, and social workers apply a questionnaire to juvenile offenders and to people who have been convicted for a crime (these are different questionnaires); the questionnaire is then processed using a model trained on thousands of previous cases, and an output is produced, which is then interpreted by the person applying the instrument. Detailed anonymized data is available for studying to what extent different instruments can be biased against different sub-populations (e.g., immigrants or children of immigrant). Additionally, we would like to explore the potential for new risk prediction instruments. The work would be done in collaboration with researchers on criminology at the University of Barcelona and at the Justice Department of Catalonia.

Carlos Castillo

Chat with your documents, table and images with Retrieval Augmented LLMs

Large language models (LLMs) and generative AI have taken the world by storm thanks to their incredible capabilities of fluency and knowledge extraction. However, often information stored in the model during training is obsolete or not relevant to the current task. One exciting possibility is to use a vector database as a memory and to let the language model access this information at inference time. This approach results in answers that are more factual answers and relevant. In this project, we will jointly investigate how to develop and validate a Retrieval Augmented system that uses LLMs to create embedding’s for multimodal documents, including text, table and images from public literature and technical reports from the European Commission.

Sergio Consoli

Multicentric clinical study on lung cancer

Lung cancer is one of the most deadly diseases in the world and recent results showed that early detection by means of an yearly Computer Tomography (CT) scan of the chest reduced mortality by 20% in a population of ex-smokers. However, once a lesion is found in the CT scan, then a biopsy is needed to confirm whether the nodule is malignant, which is a very invasive procedure. The JRC participates in a clinical study with 4 hospitals to enable and advance the use of AI in the detection and follow up of lung lesions. In the scope of this project, we will jointly work on the cloud infrastructure that processes the clinical data and on the deep learning algorithms for the analysis of the CT scans.

Valentin Comte

Reasoning with language and knowledge graphs

Large language models (LLMs), are making new waves in the field of natural language processing and artificial intelligence, due to their emergent ability and generalizability. However, LLMs are black-box models, which often fall short of capturing and accessing factual knowledge. In contrast, Knowledge Graphs (KGs), Wikipedia and Firebase for example, are structured knowledge models that explicitly store rich factual knowledge. KGs can enhance LLMs by providing external knowledge for inference and interpretability. Meanwhile, KGs are difficult to construct and evolving by nature, which challenges the existing methods in KGs to generate new facts and represent unseen knowledge. Therefore, it is complementary to unify LLMs and KGs together and simultaneously leverage their advantages. In this project, we will work jointly towards a system that unifies LLM and KGs for the processing of scientific literature and technical reports from the European Commission.

Lorenzo Bertolini

Genomic similarities via vector embeddings

Recent advances in sequencing techniques created a huge amount of public genomic data available for viruses, bacteria and mammals. That wealth of data has the potential to unlock important advances in all the fields of medicine such as communicable and non-communicable disease, pharmaceutical and antimicrobial resistance. However, processing those data in an efficient way and creating models that can generate recommendations based on them is still a challenge. In this project, we will work on deep learning methods to process genomic and molecular data to create vector embeddings that can then be used for fast similarity search.

Gabriele Leoni

Transferable Adversarial Attacks on Aligned Language Models for Health

Despite the success of deep neural networks, the existence of adversarial attacks has revealed the vulnerability of neural networks in terms of security. Adversarial attacks add subtle noise to the original example, resulting in a false prediction. Although adversarial attacks have been mainly studied in the image and speech domain, a recent line of research has discovered that LLMs are also exposed to adversarial attacks. This is now starting to appear also in the multimodality of Large Language Models. In this project we aimed, working with LLMs and making use of the research literature available, to propose multimodal adversarial attacks that can induce aligned language models to produce questionable content which does not follow the HHH principles (helpful, honest, and harmless) in the models.

Antonio Puertas Gallardo

http://www.upf.edu/web/tide

Data Analysis in Education Technologies

Several topics related to the application of AI and data analytics techniques to the design of systems that support teaching and learning (includes community, teaching and learning analytics, interactive dashboards, etc.)

Specific Topics

Davinia Hernández-Leo

Urban data governance. Beyond open data

How data from public sector, private companies and citizens can be managed and governed to extract more richness while keeping privacy and securing data explotation models. The EU’s Data Governance Act seeks for promoting data reuse and the GAIA-X project is meant to provide an architecture for such ideal. Current debates around Data Spaces and Data Trusts, some include a Federated infraestructure and some others trust in DAOs (blockchain) to provide a solution that includes democratic approach to data sharing. This TFM will explore different opportunities and question what are the assumptions behind each model and understand how different infraestructure is needed to accomplish the needs of different uses of data in the urban context. It will be focused on theoretical analysis of different models and technical assessments of current data technologies and architectures. In addition, the analysis of different spatial-temporal datasets will be analysed to study how cities can integrate and benefit from this approach.

Manuel Portela

Intersectional Fairness

During the last years, Fairness Accountability and Transparency (FaccT) has been discussed to reduce social harm and discrimination in the design and development of algorithms. Although, most of the works on FaccT approaches only address discrimination towards limited categories in just one dimension (e.g., gender, race). Recently, voices from black feminism and social studies have called to a more depth approach to a fair and just development. Intersectionality is an analytical framework for understanding how aspects of a person's social and political identities combine to create different modes of discrimination and privilege. It identifies multiple factors of advantage and disadvantage. Examples of these factors include gender, caste, sex, race, class, sexuality, religion, disability, physical appearance, and height. That is, it goes behind the classical gender/race binaries. Thus, looking at one dimension at a time doesn’t always tell us the whole story. Despite that the call for an intersectional approach is growing among the Computer Science and Data Science communities, few works explore the real possibility of working the concept of FAccT over algorithms. The goal of this thesis is to develop an exploration of diverse datasets and algorithms used in the context of social interest (e.g., social welfare, recidivism, education) and to develop technical solutions taking into account an intersectional perspective. This thesis is approached with a highly theoretical conceptualization as well as a technical development.

Manuel Portela

Fair housing algorithm

Housing market in Barcelona is under constant scrutiny due to high demand and current economic-crisis. Moreover, the housing market prices are determined by the same longstanding valuations without reflecting current societal and environmental needs (e.g. green areas, low pollution, and diversity). The goal of this thesis is to understand which factors related to climate change and pollution could affect the real state market and housing policies in the near future. The project is meant to analyze different sources of data and to analyze the current status of Barcelona’s housing market from a data-driven approach. As a result, it is expected to develop an algorithm to re-calculate the cost of housing, not under market demand-offer dynamics, but under Sustainable Development Goals (SGDs) indicators, the impact of climate change and social justice principles.

Manuel Portela

Ranking Attributes for Information Gain in Forests of Decision Trees (data science/machine learning related project)

Decision trees are one of the most well-known techniques in machine learning, data science, analytics and data mining. They are transparent as they can be presented as rules for human understanding. Explainable AI, human-in-the-loop developments, and fair and ethical AI favour the transparency of decision trees. With the availability of big data, algorithms for constructing decision trees in a platform like MapReduce are a new challenge. This project consists of developing the parallelisation algorithms and their implementation for Hadoop when we consider a forest of decision trees. This implementation will be focused on the applications of the forest of decision trees on privacy-preserving data mining on online social networks (OLSN). We can provide references to recent literature on how the construction of a forest of decision trees is used to provide privacy alerts to users of OLSN. Another application is feature selection.

Vladimir Estivill-Castro

Ethical Ai Agent

The aim is to develop a demonstration of interaction by a human being with an artificial agent about ethical dilemmas and reason over ethical challenges. The agent shall apply several techniques to argue and explain ethical decisions, including game theory or dialogue with humans. Another possibility is deontic logic. The goal is to investigate practical debates between and artificial agents and humans regarding the alignment problem. Possibilities are to explore the use of generative AI and in particular, large language models.

Vladimir Estivill-Castro

Verifiable AI

The goal is to explore mechanisms of formal verification of agents' behaviour, particularly agents who can combine reasoning with operating over a state machine. Moreover, if the behaviour or the knowledge for reasoning is learned, behavioural properties should be verified regarding the behaviour. The aim is to create trustable AI systems because they have been formally verified even if they evolve by learning. The immediate applications are behaviours of autonomous vehicles, and particular aspects could be efficient negotiation of intersections. There is already evidence that artificial agents can negotiate intersections faster than humans and even minimise the requirement of traffic signals. However, there are challenges is mixed environments, such as humans and autonomous vehicles. The impact on mobility of artificial intelligence tools is the focus of this project.

Vladimir Estivill-Castro

Statistical Modeling of Online Discussions

Online discussion is a core feature of numerous social media platforms and has attracted increasing attention for different and relevant reasons, e.g., the resolution of problems in collaborative editing, question answering and e-learning platforms, the response of online communities to news events, online political and civic participation, etc. This project aims to address, from a probabilistic modeling perspective, some existing challenges, both computational and social, that appear in platforms that enable online discussion. For example, how to deal with scalability issues, how to evaluate and improve the quality of online discussions, or how to generally improve social interaction such platforms.

Generative models of online discussion threads: state of the art and research challenges

Abstraction and Reasoning Challenge (Kaggle)

Andreas Kaltenbrunner

Abstraction and Reasoning Challenge

This project is about addressing the Abstraction and Reasoning Challenge (ARC), a kaggle competition consisting of large corpus of visual tasks that require abstraction and reasoning skills to be solved. Each ARC task contains 3-5 pairs of train inputs and outputs, and a test input for which you need to predict the corresponding output with the pattern learned from the train examples. The objective of this project to apply AI techniques (and possibly suggest novel ones) that combine search and learning for automated program synthesis that are developed in the AI/ML research group to address the challenge, or a part of it.

[1] Mikolov et al. Efficient Estimation of Word Representations in Vector Space, 2013

Solving matching and alignment tasks with Neural Networks

In this project you will be working in the project Matching Learning (red.es BMATUPF) in close collaboration with BMAT. BMAT is a music company that uses artificial intelligence to index, monitor and report music usage and ownership data across TVs, radios, venues and digital platforms worldwide. This offer gives the opportunity to work on a highly new and innovative field: the application of state of the art Natural Language Processing tools based on Deep Learning and Large Language Models (GPTx and open source versions of ChatGPT) to process (semi and unstructured) data.. The challenges that you will be addressing will be realistic and relevant for industry as they will be provided directly from BMAT. The research will be centered on the power of mixing Transformers [1,2] and pretrained transformers (like ChatGPT), Graph Neural Networks and Knowledge Graphs for having higher capabilities when solving matching and entity alignment tasks [3,4]. A good level of Python and some experience with Neural Network libraries (keras, tensorflow or pytorch) would be needed. Python notebook will be used.

BMAT

[2] M Leone et al. Attention is all you need

[3] N Fanourakis et al. A Critical Re-evaluation of Neural Methods for Entity Align

[4] Knowledge graph embedding methods for entity alignment: experimental review. Data Mining and Knowledge Discovery (2023) 37:2070–2137

Martí Sanchez-Fibla

Graph Neural Networks for Large-scale Optimization

Large-scale constrained optimization problems are frequently encountered in real-world applications, e.g. traffic management, logistic planning, warehouse optimization, multiagent planning, and scheduling, to name a few. Though constrained optimization problems have been extensively studied, because of their computational complexity, existing optimal solvers do not scale well to complex problems. Recently, graph neural networks (GNNs) have been successfully applied to approximate solutions to large optimization problems. The aim of the thesis is to study the potential of graph neural networks for large-scale optimization, and test it in realistic applications.

Anders Jonsson

Linearly-solvable Markov decision processes with function approximation

A linearly-solvable Markov decision process (LMDP) is a special case of the more commonly used Markov decision process (MDP) in reinforcement learning. In LMDPs, the Bellman equations governing the value function are linear, which typically makes learning more efficient. In spite of being simpler than MDPs, LMDPs are surprisingly expressive and can be used to represent many reinforcement learning benchmarks. In particular, any MDP with deterministic actions can be converted into an LMDP. However, the large body of work on deep reinforcement learning largely ignores LMDPs as a representation. The goal of this project is to develop function approximation algorithms for LMDPs, e.g. deep neural networks. One way to do so is to study existing algorithms for MDPs and simplify them for the LMDP setting..

Anders Jonsson

A unified view of entropy-regularized Markov decision processes

Entropy-regularized reinforcement learning algorithms

Entropy regularization is an extremely popular component of modern reinforcement learning methods. After several years of applying entropy regularization as a heuristic to drive exploration and improve optimization properties of RL algorithms, there has been lots of recent progress in providing solid theoretical foundations justifying this technique. Most recently, a sophisticated regularization technique has lead to the development of a principled RL algorithm, Q-REPS, that is entirely derived from first principles, yet can be implemented in large-scale environments essentially without any approximations to the theory. This project aims to progress the research on this method from two parallel directions: 1) Implement Q-REPS in large-scale environments and analyze its performance empirically and 2) analyze various properties of the algorithm that remain poorly understood. (Some important questions being: understanding the role of the regularization functions and finding alternatives, or studying the properties of the action-value functions returned by the algorithm.) The project requires strong mathematical skills, particularly in linear algebra and probability theory. Knowledge of convex analysis and optimization is a plus, but not absolutely necessary at the current stage.

Logistic Q-Learning

Approximate Modified Policy Iteration and its Application to the Game of Tetris

Error averaging in approximate dynamic programming

Error-propagation analysis of approximate dynamic programming methods is a classic research topic with several implications on reinforcement learning algorithms. The most well-known results consider basic DP algorithms like value iteration and policy iteration, with several new results added in recent years regarding regularized counterparts of these methods. One discovery on this front was the discovery that certain regularization choices result in an "error-averaging" property that ensures a benign propagation of policy evaluation errors: roughly speaking, instead of accumulating the absolute errors linearly, these schemes allow the cancellation of positive and negative errors. This project sets out to understand this phenomenon in more detail by decoupling the averaging effect from the regularization effects as much as possible, thus potentially enabling the development and analysis of more effective error-averaging RL algorithms. The project requires strong mathematical skills, particularly in linear algebra and probability theory. Knowledge of convex analysis and optimization is a plus, but not absolutely necessary at the current stage.

Leverage the Average: an Analysis of KL Regularization in RL

Munchausen Reinforcement Learning

Learning to Optimize Via Posterior Sampling

Thompson sampling for sequential prediction

Thompson sampling is one of the most well-studied algorithms for a class of sequential decision-making problems known as stochastic multi-armed bandit problems. In a stochastic multi-armed bandit problem, a learner selects actions in a sequential fashion, and receives a sequence of rewards corresponding to the chosen actions. A crucial assumption made in this problem is that the rewards associated with each action are random variables drawn independently from a fixed (but unknown) distribution. The goal of this project is to do away with this assumption and study Thompson sampling in non-stationary environments where the rewards may be generated by an arbitrary external process. Precisely, the project considers the framework of sequential prediction with expert advice, and aims to show theoretical performance guarantees for this algorithm and/or analyze its perormance empirically. The project requires strong mathematical skills, particularly in probability theory and multivariate calculus. Knowledge of convex analysis is a plus, but not absolutely necessary at the current stage.

An Information-Theoretic Analysis of Thompson Sampling

Online Linear Optimization via Smoothing

Thompson Sampling for Adversarial Bit Prediction

Online Linear Quadratic Control

Better algorithms for online linear-quadratic control

Linear-quadratic control is one of the most well-studied problem settings in optimal control theory: it considers control systems where states follow a linear dynamics, and the incurred costs are quadratic in the states and control inputs. In recent years, the problem of online learning in linear-quadratic control problems have received significant attention within the machine-learning community. One particularly interesting development is the formulation of the control problem as a semidefinite program (SDP), which allows the application of tools from online convex optimization. The present project aims to develop new algorithms for online linear quadratic control based on this framework by exploring the possibility of using regularization functions that make better use of the SDP geometry than existing methods based on online gradient descent. The project requires strong mathematical skills, particularly in multivariate calculus and linear algebra. Knowledge of control theory, convex analysis and optimization is a plus, but not absolutely necessary at the current stage.

Online PCA with Optimal Regret