Luis Roque
LIACC / FEUP

Abstract
Evaluating time series models creates unique challenges, especially as the field advances toward large-scale foundation models. Traditional evaluation pipelines often overlook how dataset selection can introduce significant bias, leading to overly optimistic or misleading conclusions. In this talk, we will examine the critical role of dataset curation in shaping evaluation outcomes and explore strategies for mitigating bias through augmentation and semi-synthetic data generation. Beyond improving evaluation robustness, we will introduce new methodologies designed to assess resilience in scenarios where transfer learning across domains is essential. By connecting practical evaluation techniques with the emerging paradigm of time series foundation models, we aim to provide a forward-looking framework that enables researchers and practitioners to conduct more reliable and generalizable assessments.

2025 October 24, 13:30

IA/U.Porto
Centro de Astrofísica da Universidade do Porto (Auditorium)
Rua das Estrelas, 4150-762 Porto