Low-rank Methods in Visual Computing and Machine Learning
Speaker: Rafael Ballester-Ripoll, IE School of Science and Technology (Madrid)
Low-rank decompositions of matrices and tensors periodically return to the spotlight in scientific computing and data science: dimensionality reduction, signal recovery, recommender systems, data compression, efficient LLM finetuning, etc. In this talk, I will overview low-rank methods with a focus on higher dimensions (tensor decompositions) and their applications in visual data: analysis of computer tomography scans, compression of physical simulations and other large-scale datasets, and scientific visualization of mathematical models in the engineering sciences. We will also dive into new exciting developments in integrating low-rank tensor methods into deep learning pipelines, thus opening promising novel venues for scalable and explainable artificial intelligence (XAI).