Student Presentations Series

In the Student Presentations Series, ELLIS Alicante invites PhD students to present their work to the team of ELLIS Alicante. In collaboration with the University of Alicante, the event is open to any student and researcher of the university.

Talks

  • On the Relationship of Fairness and Uncertainty

    Abstract: To ensure the reliability and trustworthiness of machine learning models in real-world applications, it's crucial to assess the uncertainty about their predictions (predictive uncertainty). In a practical scenario, uncertain decisions will be deferred to human experts or opted for a failsafe alternative. Machine learning models already operational in our society to make inferences or predictions in consequential domains in people's lives ---such as healthcare, access to credit, social programs or promotions--- have been found to exhibit differences in their performance depending on the person's gender, race, or age, leading to unfair treatment to certain groups of individuals. Fairness research aims to identify, analyze, and address these systematic biases in machine learning models. This presentation will delve into the interplay between predictive uncertainty and algorithmic fairness, addressing questions such as: Do models exhibit the same levels of unfairness at high or low certainty levels? What kind of uncertainty leads to unfairness? Is it possible to leverage the predictive uncertainty to develop fairer models? I will provide a literature review of current knowledge about those questions, present our latest experimental insights, and discuss future research avenues.

    Short bio: Kajetan Schweighofer is an ELLIS PhD student and holds a bachelor’s and master’s degree in physics as well as a second master’s degree in artificial intelligence from Johannes Kepler University Linz. The master’s thesis in artificial intelligence supervised by Dr. Sepp Hochreiter focused on Offline Reinforcement Learning. The master’s thesis in physics supervised by Dr. Armando Rastelli focused on utilizing artificial intelligence for quantum optical experiments. Kajetan conducts his PhD under the supervision of Dr. Sepp Hochreiter at Johannes Kepler University Linz and Dr. Nuria Oliver at ELLIS Alicante. He focuses on improving uncertainty quantification for deep learning methods, to make those methods reliable and trustworthy when applied in critical settings. Furthermore, during the exchange in Alicante throughout the winter semester 2023/24, Kajetan investigates implications of uncertainty on the fairness of deep learning methods.

    Presenter: Kajetan Scheweighofer

    Date: 2024-03-26 12:00 (CET)

    Location: Oficinas ELLIS Alicante, Muelle Pte., 5 – Edificio A, Alicante 03001, Alicante ES

  • Towards Student-Centric AI-Supported Learning: Teaching Chatbots to Ask the Right Questions

    Abstract: In the rapidly evolving landscape of educational technology, the integration of Large Language Model (LLM)-based chatbots presents a transformative opportunity to enhance learning experiences. Our project introduces "Maike", an innovative educational chatbot designed to revolutionize the traditional learning paradigm by fostering critical thinking, purposeful learning, and self-efficacy among students. Unlike conventional chatbots that provide immediate answers, Maike employs Socratic dialogues to engage students, prompting them to explore and reflect upon their queries through guided questions. During this presentation, we will delve into a comprehensive literature review highlighting the significance of such technologies in education and detail the theoretical framework underpinning Maike.

    Short bio: Lucile Favero is an ELLIS PhD student, with a bachelor’s degree in mathematics from the University of Geneva and a master’s degree in Mathematics and Neuroscience. In her master thesis in Mathematics, directed by Dr. Sylvain Sardy, she developed a machine learning model to optimize heat pump systems. In parallel, in her neuroscience studies, Lucile did a full-time internship for three years under the supervision of Dr. Giulio Matteucci at the El-Boustani laboratory. There, she implemented a decision-making model that promises to provide invaluable information to experimenters when designing, analyzing and refining behavioral tasks. Lucile is currently pursuing her PhD studies under the guidance of Dr. Nuria Oliver (ELLIS Alicante), Dr. Tanja Kaser (EPFL) and Dr. Juan Antonio Pérez Ortiz (UA) focusing on the integration of the Socratic method in chatbot development to improve educational outcomes.

    Presenter: Lucile Favero

    Date: 2024-03-01 11:30 (CET)

    Location: Salon de Actos Politecnica IV, Carretera San Vicente del Raspeig s/n, San Vicente del Raspeig 03690, Alicante ES

    Online: https://vertice.cpd.ua.es/291498

  • Navigating Privacy in Healthcare tech for Senior Care, User Side of the Story

    Abstract: Active and Assisted Living (AAL) technologies strive to enrich the quality of life of older adults and facilitate successful aging. While video-based AAL solutions offer rich capabilities for better healthcare management in older age, they pose significant privacy risks. In response to these challenges, the AMI4AHA (Ambient Intelligence for Active and Healthy Ageing) research team at the University of Alicante has developed a video-based monitoring system for senior care. The system incorporates various privacy-preserving filters to address the privacy risks effectively. The presenter aims to offer insights from the user perspective regarding such technologies. She will delve into her three-year research process, which focuses on comprehending the diverse perspectives of stakeholders involved in the caregiving and receiving process.

    Short bio: Tamara Mujirishvili is an Early Stage researcher at the Marie Curie ITN VisuAAL, based at the university of Alicante. Her research project focuses on the perceptions of personal safety and privacy in older adults in the context of video-based lifelogging technologies. Tamara received her Master's Degree in Neuroscience from the University of Bordeaux (France) in 2020. She executed her master thesis research at the University of Cambridge investigating how prior expectations influence perception. She obtained her BA Degree in Psychology from Tbilisi State University (Georgia) in 2015. She has spent the 2013-2014 academic year at the University of Groningen (The Netherlands) as an exchange student at the faculty of Behavioral and Social Sciences, specializing in Psychology.

    Presenter: Tamara Mujirishvili

    Date: 2024-02-22 14:00 (CET)

    Location: Distrito Digital 5, Muelle Pte., 5 – Edificio D, Alicante 03001, Alicante ES

  • Probability and Machine Learning

    Abstract: In this talk, I will challenge common ideas about probability, particularly in the context of machine learning. The main point is that there are no correct probabilities, that they are always constructed rather than discovered. I will show that constructed probabilities can still be useful and specify the assumptions under which e.g. expected utility maximisation is a sensible policy (which is usually taken for granted). Based on these general considerations, I will outline some insights for algorithmic fairness. As for other parts of my talk, central aspects are the tension between individuals and groups as well as a general notion of calibration. I will then draw on previously discussed issues to criticise the way individual predictions are typically presented in machine learning. In the last part, I will argue against the true distribution framework that underlies almost every analysis of theoretical guarantees in machine learning.

    Short bio: Ben holds Bachelor’s degrees in Mathematics and Philosophy from LMU Munich as well as Master’s degrees in Philosophy of Science from LMU and in Computer Science from the University of Oxford. He continued to work there as a research assistant with the OATML group before starting a PhD at the University of Tübingen with Bob Williamson. Ben is co-supervised by Nuria Oliver within the ELLIS programme and will spend the winter semester 2024/25 in Alicante. His main interests lie at the intersection between the social impact of algorithmic predictions and the foundations of machine learning and probability.

    Presenter: Benedikt Höltgen

    Date: 2024-02-14 11:00 (CET)

    Location: Oficinas ELLIS Alicante, Muelle Pte., 5 – Edificio A, Alicante 03001, Alicante ES

  • Variational Mixture of HyperGenerators for learning distributions over functions

    Abstract: Recent approaches build on implicit neural representations (INRs) to propose generative models over function spaces. However, they are computationally costly when dealing inference tasks, such as missing data imputation, or directly cannot tackle them. In this presentation, we will talk about a novel deep generative model, Variational Mixture of HyperGenerators (VAMoH). VAMoH combines the capabilities of modeling continuous functions using INRs and the inference capabilities of Variational Autoencoders (VAEs). Through experiments on a diverse range of data types, such as images, voxels, and climate data, we show that VAMoH can effectively learn rich distributions over continuous functions. Furthermore, it can perform inference-related tasks, such as conditional super-resolution generation and in-painting, as well or better than previous approaches, while being less computationally demanding.

    Short bio: Batuhan Koyuncu is an ELLIS PhD student at Saarland University, advised by Isabel Valera and co-advised by Ole Winther. His research interests include building expressive, efficient, and interpretable deep generative models, and utilizing their applications in psychiatry and healthcare.

    Presenter: Batuhan Koyuncu

    Date: 2023-04-05 10:00 (CEST)

    Location: Salon de Actos Politecnica IV, Carretera San Vicente del Raspeig s/n, San Vicente del Raspeig 03690, Alicante ES

    Online: https://vertice.cpd.ua.es/281053

  • Towards Explainable AI

    Abstract: AI systems have become the de facto tools to solve complex problems in computer vision. Yet, it has been shown that these systems might not actually be safe to be deployed in the real world, as they too often tend to rely on dataset biases and other statistical shortcuts to achieve high performance. A growing body of research thus focuses on the development of explainability methods to better interpret these systems, to make them more trustworthy. In this talk, I will first give a general overview of the methods commonly used in eXplainable AI, before discussing the challenges that still need to be overcome by the community.

    Short bio: Julien Colin is an ELLIS PhD student. He holds a Bachelor’s degree in Physics and Chemistry (2019, University of Lorraine) and a Master’s degree in Cognitive Sciences: Natural and Artificial Cognition (2021, INP Grenoble). Before the start of his PhD, he worked as a research assistant; first at ANITI for 6 months (2021, Toulouse) then at Brown University for 5 months (2022, Providence). His PhD topic is centered around eXplainable AI and Cognitive Science. In his research, he is interested in developing methods to better understand Intelligent systems. His supervisors are Nuria Oliver (ELLIS Alicante) and Thomas Serre (ANITI).

    Presenter: Julien Colin

    Date: 2023-01-30 11:30 (CET)

    Location: Salon de Actos Politecnica IV, Carretera San Vicente del Raspeig s/n, San Vicente del Raspeig 03690, Alicante ES

    Online: https://vertice.cpd.ua.es/278153

  • Modeling Irregular Time Series with Continuous Recurrent Units

    Abstract: Recurrent neural networks (RNNs) are a popular choice for modeling sequential data. Modern RNN architectures assume constant time-intervals between observations. However, in many datasets (e.g. medical records) observation times are irregular and can carry important information. To address this challenge, we propose continuous recurrent units (CRUs) -- a neural architecture that can naturally handle irregular intervals between observations. The CRU assumes a hidden state, which evolves according to a linear stochastic differential equation and is integrated into an encoder-decoder framework. The recursive computations of the CRU can be derived using the continuous-discrete Kalman filter and are in closed form. The resulting recurrent architecture has temporal continuity between hidden states and a gating mechanism that can optimally integrate noisy observations. We derive an efficient parameterization scheme for the CRU that leads to a fast implementation f-CRU. We empirically study the CRU on a number of challenging datasets and find that it can interpolate irregular time series better than methods based on neural ordinary differential equations.

    Short bio: Mona Schirmer is an ELLIS PhD student at the University of Amsterdam and a machine learning consultant at the World Bank. Before that, she completed a master’s in statistics at Humboldt University and Technical University in Berlin as well as a French engineering diploma at ENSAE. In her Bachelor’s, she studied economics and political science at Humboldt University of Berlin and the University of Munich. Her research interests lie in probabilistic machine learning und machine learning for social good. She visited ELLIS Alicante in May 2022.

    Presenter: Mona Schirmer

    Date: 2022-05-25 11:30 (CEST)

    Location: Salon de Actos Politecnica I, Carretera San Vicente del Raspeig s/n, San Vicente del Raspeig 03690, Alicante ES

    Online: https://vertice.cpd.ua.es/269014

  • AI and Human Biases

    Abstract: Picture this - you are at the movies waiting in line to buy popcorn. On the big, bright screen behind the counter you see multiple options. You take your time and carefully crunch all the numbers in front of you and order what you think is the optimal choice. You walk away with a big smile on your face because you just got a very good deal. However, chances are you just got tricked into buying something you didn’t even want to begin with. This is a classic example of sub-optimal human decision making that has been studied in behavioral economics. However, the penetration of these ideas into Artificial Intelligence has been fairly limited. Given that in today's world humans and machines work closely together, it’s important to design systems that can account for the fact that our decisions are heavily influenced by multiple cognitive biases. This talk is intended to be an introduction to the design of such systems. We will look at multiple biases and how they have been used in the real world. We will also look at work in Artificial Intelligence that tries to account for these biases and propose future directions for this area.

    Short bio: Aditya Gulati is an ELLIS PhD student. He holds a Bachelor’s and Master’s degree in Computer Science Engineering from the International Institute of Information Technology Bangalore. His research interests lie in modelling human behavior and artificial intelligence. His supervisors are Nuria Oliver (ELLIS Alicante), Miguel Angel Lozano (University of Alicante) and Bruno Lepri (Fondazione Bruno Kessler).

    Presenter: Aditya Gulati

    Date: 2022-02-28 11:30 (CET)

    Location: Salon de Actos Politecnica IV, Carretera San Vicente del Raspeig s/n, San Vicente del Raspeig 03690, Alicante ES

    Online: https://vertice.cpd.ua.es/263593

  • AI on social media: cultural impact and proposed solutions

    Abstract: Social media platforms have an unprecedented impact on our culture, exacerbating existing social disparities and filtering our vision of the world. Unlike previous media, these platforms offer enormous possibilities to carry out analytical research. Given the wide availability of data, we can deepen our understanding of inequalities and take concrete actions to mitigate them. In this talk, I will present cultural paradigms that create detriment for the life of women, both online and offline. In particular, I will focus on the role of Artificial Intelligence, and specifically the impact that autonomous algorithms -pervasive on social media- have on women. I will propose a proactive research perspective to achieve positive cultural impact with AI.

    Short bio: Piera Riccio is an ELLIS PhD student. She holds a bachelor’s degree in Cinema and Media Engineering (2018, Politecnico di Torino), a Master’s degree in ICT for Smart Societies (2021, Politecnico di Torino), and a Master’s degree in Data Science and Engineering (2021, Télécom Paris – EURECOM). In 2020, she was an affiliate at Metalab (at) Harvard. In 2021, she was a research assistant at the Oslo Metropolitan University. In her research, she is interested in exploring the cultural, social, and artistic possibilities of AI. In her PhD, she focuses on the effect that social media have on the lives of women and the way they are perceived in the social media cultural ecosystem. Her supervisors are Nuria Oliver (ELLIS Alicante), Thomas Hofmann (ETH Zurich) and Francisco Escolano (Universidad de Alicante).

    Presenter: Piera Ricco

    Date: 2022-02-14 11:30 (CET)

    Location: Salon de Grados Derecho, Carretera San Vicente del Raspeig s/n, San Vicente del Raspeig 03690, Alicante ES

    Online: https://vertice.cpd.ua.es/263873

  • Fairness in Federated Learning: Challenges and Opportunities

    Abstract: Federated Learning is a popular, emergent field within Machine Learning. It has advantages and challenges. It can help implement Machine Learning in tasks with low, sparse data sources or without strong central computing power. It can be a good answer for clients who do not want to share their data with the rest of the world, yet they want to benefit from the generalized performance of a global model. However, federated learning models can suffer from bias. Most of the current fairness-ensuring methods in machine learning use the relative position of the individuals within the user space to ensure the model's fairness when compared to the rest of the users. Unfortunately, it is not trivial to apply such an approach in a federated architecture. For example, how can we identify bias against a certain group (e.g. race, sex) if we don't have a global evaluation of fairness metrics? Investigating the trade-offs between privacy and fairness is an open question in Federated Learning that I will investigate in my PhD thesis. This talk provides a general overview of Federated Learning research and shows the unique challenges and solutions of fairness guarantees in a federated learning setup.

    Short bio: Gergely Dániel Németh is a PhD student at ELLIS Alicante. His PhD topics are AI Ethics and Federated Learning. His supervisors are Nuria Oliver (ELLIS Alicante), Miguel Angel Lozano (University of Alicante) and Novi Quadrianto (University of Sussex). He holds a CS MSc degree from The University of Manchester and a BSc from Budapest University of Technology and Economics. His university topic was about Natural Language Processing, but he also worked on Computer Vision in a Hungarian StartUp.

    Presenter: Gergely D Németh

    Date: 2022-01-31 11:30 (CET)

    Location: Salon de Actos Politecnica I, Carretera San Vicente del Raspeig s/n, San Vicente del Raspeig 03690, Alicante ES

    Online: https://vertice.cpd.ua.es/261653

  • Fairness in Machine Learning

    Abstract: Machine learning models are becoming the main tools for addressing complex societal problems and are also increasingly deployed to make or support decisions about individuals in many consequential areas of their lives, from justice to healthcare. Therefore, the ethical implications of such decisions, including concepts such as privacy, transparency, accountability, reliability, autonomy, and fairness need to be taken into account. Specifically, we will explain the current landscape in AI Fairness, from the sources of the bias and different algorithmic fairness approaches to their limitations and cutting-edge approaches. The main goal is to provide a general overview of what is Fairness as well as the main research challenges that the community has to address.

    Short bio: Adrián Arnaiz Rodríguez is an ELLIS PhD student. He holds a Bachelor’s degree in Computer Engineering (2019, Universidad de Burgos) and a Master’s degree in Data Science and Artificial Intelligence (2021, Universitat Oberta de Catalunya) doing the MSc thesis with Baris Kanber (University College London) in medical neuroimaging. His PhD supervisors are Nuria Oliver (ELLIS Alicante), Francisco Escolano (Universidad de Alicante) and Manuel Gómez Rodríguez (Max Planck Institute for Software Systems). His PhD topics are AI Fairness, Causality and Graph Theory to enhance ethics, accountability, and transparency in algorithmic decision-making.

    Presenter: Adrián Arnaiz Rodríguez

    Date: 2022-01-17 10:00 (CET)

    Location: Salon de Actos Politecnica IV, Carretera San Vicente del Raspeig s/n, San Vicente del Raspeig 03690, Alicante ES

  • Data-efficient methods for model learning and control

    Abstract: In this talk, I will present an overview of the research I have been working on during my Ph.D. project, focused on data-efficient methods in machine learning. The first part will be dedicated to learning models of dynamic systems. Models facilitate simulations, analysis of the system's behavior, decision making, and design of automatic control algorithms. Even inherently model-free control techniques such as reinforcement learning have been shown to benefit from the use of models. However, obtaining informative data for constructing dynamic models can be difficult, especially when the models are to be learned during task execution. To this end, symbolic regression proves to be a suitable method to automatically build such models. This technique, based on genetic programming, constructs from data parsimonious models in the form of analytic equations. It represents an alternative to the currently popular data-hungry deep learning methods, which typically produce black-box models. One of the challenges in continual model learning is posed by the large amount of data collected from the system. I will present a comparison of methods for selecting informative training samples and show that symbolic regression can be used to construct accurate models from very small informative data sets. Furthermore, I will explain how symbolic regression can be naturally extended to account for physical constraints and a partially known theoretical or empirical model of the system. However, data-efficient methods are needed not only to learn models of the dynamic systems but also to learn models of the environment. Therefore, the talk will be concluded by presenting a method from the field of computer vision for robotics. Ubiquitous changes of the scene structure and appearance are typical for real-world dynamic environments and make many conventional localization and navigation methods fail. I will present a method for change detection based on weighted local visual features that improves the localization accuracy by distinguishing between stable parts of the scene and potentially confusing changing regions.

    Short bio: Erik Derner received the M.Sc. degree in artificial intelligence and computer vision from the Czech Technical University (CTU) in Prague, where he is currently pursuing a Ph.D. degree. Throughout his studies, he acquired a broad international experience during five stays at universities in four different countries, ranging from a few months to the full academic year. His research interests are focused on sample-efficient methods for model learning applied to dynamic systems such as mobile robots and their environments. The main areas of research comprise genetic programming, reinforcement learning, and computer vision. The central topic in his research is the use of symbolic regression to automatically construct nonlinear models of dynamic systems.

    Presenter: Erik Derner

    Date: 2021-12-13 12:00 (CET)

    Location: Laboratorio de Grados Politecnica I, Carretera San Vicente del Raspeig s/n, San Vicente del Raspeig 03690, Alicante ES