Research seminars

Bio: I'm a Full Professor in the Computer Science Department at Universidad de Zaragoza, where I lead the Graphics & Imaging Lab of the I3A Institute. I'm also a member of the Vision, Image and Neurodevelopment Group of the IIS Aragon Institute. I'm the recipient of the 2022 Eurographics Outstanding Technical Contributions Award. My research focuses on the areas of rendering (simulation of light transport), computational imaging, perception, and virtual reality. I have been a visiting researcher at MIT, Stanford, Yale and UCSD, among others, and was recently selected as one of the 100 most influential researchers of the decade in his field. 

Abstract: Although optical design is a mature field, the introduction of novel optical devices such as metasurfaces will require a concurrent introduction of new design methods. Coincident with the invention of these new light-shaping tools is the rise of artificial intelligence, specifically deep learning with neural networks. In this talk, I will present my research on differentiable wave propagation and its application to cameras and displays. Specifically, the optical components are treated as differentiable layers, akin to neural network layers, that can be trained jointly with the computational blocks of the imaging/display system. I will show how this framework can be used to design salt-sized metasurface optics, commercial camera optics, and étendue expanding optics for holographic displays.

Bio: Ethan Tseng is a founder and CTO of Cephia, a company that aims to redefine computer vision and computational imaging. He received his PhD from Princeton University where he was advised by Prof. Felix Heide. Ethan’s research was highlighted by Optics & Photonics News in 2021 and in 2024 and has been featured in international media such as BBC, NSF Discovery Files, Newsweek, Nvidia Technical Blog, and Jimmy Fallon’s Tonight Show. Ethan is a recipient of the Google PhD Fellowship. Webpage: https://ethan-tseng.github.io 

Bio: Lorenzo Mur-Labadia is currently a researcher at Multiverse Computing. He obtained his PhD in Deep Learning and Computer Vision at the University of Zaragoza (Spain), supervised by Prof. Rubén Martínez-Cantín and Prof. Josechu Guerrero. His research focuses on video understanding, video–language learning, and 4D scene representations; with applications to embodied and egocentric perception. In the last months of his PhD, Lorenzo was a Research Scientist intern at Meta AI (FAIR) in Paris, where he worked with Adrien Bardes and Yann LeCun on large-scale self-supervised video understanding.

Abstract: Lorenzo will present his recent work, V-JEPA 2.1, a family of self-supervised models that learn dense, high-quality visual representations for both images and videos while retaining strong global scene understanding. The approach combines four key components. First, a dense predictive loss uses a masking-based objective in which both visible and masked tokens contribute to the training signal, encouraging explicit spatial and temporal grounding. Second, deep self-supervision applies the self-supervised objective hierarchically across multiple intermediate encoder layers to improve representation quality. Third, multi-modal tokenizers enable unified training across images and videos. Finally, the model benefits from effective scaling in both model capacity and training data. Together, these design choices produce representations that are spatially structured, semantically coherent, and temporally consistent.
Empirically, V-JEPA 2.1 achieves state-of-the-art performance on several challenging benchmarks, including 7.71 mAP on Ego4D for short-term object-interaction anticipation and 40.8 Recall@5 on EPIC-KITCHENS for high-level action anticipation, as well as a 20-point improvement in real-robot grasping success rate over V-JEPA-2 AC. The model also demonstrates strong performance in robotic navigation (5.687 ATE on TartanDrive), depth estimation (0.307 RMSE on NYUv2 with a linear probe), and global recognition (77.7 on Something-Something-V2). These results show that V-JEPA 2.1 significantly advances the state of the art in dense visual understanding and world modeling.