Na próxima 4ª feira, dia 8 de outubro, teremos duas talks no âmbito do processo de scouting.

Autor e vínculos: Ruxandra Barbulescu, INESC-ID Lisboa

Bio: Ruxandra Barbulescu is an auxiliary (R2) researcher and Executive Coordinator of one of the four research lines at INESC-ID, Life and Health Technology, where she leads the development of the research line and supports the development of large multi-disciplinary projects. With a background in computer science, she finished a PhD in Electrical Engineering and Computational Neuroscience at the Laboratory of Numerical Modeling, Politehnica University of Bucharest, Romania.

With 5+ years of teaching experience as Assistant Professor, she has been working as a researcher at INESC-ID in various projects on topics related to algorithms for efficient simulation of biological neural networks and responsible AI models for health. Her research is inter-disciplinary, at the intersection of modelling biological systems, numerical methods, and computer science, currently focusing on Scientific Machine Learning for improved models in terms of accuracy, interpretability and robustness.

Título: From equations to data: Towards more reliable models with Scientific Machine Learning

Resumo: In this talk, I will reflect on my research experience on physics-based as well as data-based modelling, which led me to my current research goal, of combining physical modelling based on first principles with the versatility of data-driven machine learning and the power of model reduction into hybrid Scientific Machine Learning models. I will discuss how such integration can improve the accuracy, interpretability, robustness, and consequently the reliability of scientific machine learning models while simultaneously reducing data requirements and accelerating model training. By incorporating physically-valid constraints into the learning process and integrating compression techniques, my goal is to generate more accurate and more realistic predictions for real-world applications, namely dynamic systems and biological systems for health and medicine.

Scientific machine learning has strong links to scientific/mechanistic modelling, numerical computing, machine learning and high-performance computing, its multi-disciplinarity being able to benefit students with different backgrounds.

-

Autor e vínculos: Dr. Colin Groth, New York University

Bio: Dr. Colin Groth is a postdoctoral researcher in immersive computing at New York University, specializing in perception-driven visual computing for immersive technologies. He has particular expertise in optimization for AR/VR, perceptual modelling, and deep learning integration. He received his Doctor of Engineering with summa cum laude honors from Technische Universität Braunschweig, contributing widely published research on cybersickness mitigation, gaze-contingent interaction, and computational perception. Dr. Groth has international teaching and research experience and has received multiple awards and research grants.

Título: Perception-Aware Visual Computing for the Next Generation of Immersive Media
Resumo: My talk highlights how visual computing can empower the next generation of immersive technologies. By developing algorithms like image-space warping for cybersickness reduction or foveated wavelet coding for efficient 360 video playback, I push the boundaries of technology. Thereby, my research emphasizes theoretical and algorithmic foundations and includes machine learning for complex tasks that exceed the limits of stochastic modelling. All these developments go beyond pixels and consider the unique properties of human vision as the final receiver. For example, I found that with precise modulations in the audio signal, depth perception can be tricked to converge faster on distant targets, giving users "superhuman" capabilities. While my goal is to improve user experiences, I push the boundaries of what computer graphics algorithms can achieve by applying insights from perception, efficient computation, and machine learning.