The Center for Responsible AI, led by Unbabel, held the Center for Responsible AI Forum ’24 on November 19 at the iconic Casa da Música in Porto. For the first time, the forum brought together AI startups, research centers, and industry leaders to showcase six innovative AI products and discuss the societal impact of artificial intelligence.
The event also saw the announcement of the winners of the first edition of the SPARK Awards, which recognise the best research in Responsible AI by Masters and PhD students. Among the highlights was Cátia Teixeira, the runner-up out of 16 entries with her presentation “Hubris Benchmarking with AmbiGANs”, developed as part of her Master’s thesis in Data Science and Engineering (MECD), in collaboration with Inês Gomes (student of the Doctoral Program in Informatics Engineering (ProDEI)) and Jan van Rijn (Professor at Leiden University).
One of the major limitations of R&D in ML/AI is proper evaluation of models and algorithms. These models are taking an increasingly important role in our society and economy but evaluation still focuses on predictive accuracy, which is well-known to be insufficient. Cátia proposed AmbiGANs, a methodology that generates datasets of ambiguous instances to estimate whether predictive models are overconfident.
The forum also featured presentations by other students from the Department of Informatics Engineering, who explored new methodologies to improve the evaluation of models and algorithms:
Ricardo Inácio (Master in Informatics and Computing Engineering (M.EIC), in progress) with Meta-learning and Data Augmentation for Stress Testing Forecasting Models, presented at Discovery Science;
Inês Gomes (ProDEI student, in progress) with Finding Patterns in Ambiguity: Interpretable Stress Testing in the Decision Boundary, presented at a CVPR workshop;
Luís Roque (ProDEI student, in progress) with RHiOTS: A Framework for Evaluating Hierarchical Time Series Forecasting Algorithms, presented at KDD.
The forum featured keynote speeches from notable figures such as Virginia Dignum (UN AI Advisory Body), Pedro Bizarro (Feedzai), Francisco Pereira (National Institute of Mental Health, USA), André Martins (Unbabel), and Pedro Saleiro (Opnova).
With panels like “AI Regulation – Time to (AI) Act,” the event emphasized the need for ethical and responsible AI usage, ensuring technological advances align with positive societal impact.