About this Event
135 N Skinker Blvd, St. Louis, MO 63112, USA
#SeminarFactorized Networks for Continual, Multi-task, and Multi-Modal Learning
Abstract: I will present some of our recent work on adapting deep networks for continual, multi-task/domain, and robust multi-modal learning. Our key idea is to learn a small number of task/domain/modality-specific factors that can be combined with a base network for various applications. For instance, we can learn depth estimation and segmentation using a single shared network and a small number of task-specific low-dimensional factors. We can combine data from multiple sources using different factorized and shared modules to learn features for downstream tasks. Furthermore, we can learn multiple tasks or modality combinations jointly or sequentially in a parameter-efficient manner without catastrophic forgetting. More details can be found in these papers (arXiv: 2207.09074, 2310.03986, 2310.06124, 2310.06235).
Event Details
See Who Is Interested
+ 1 People interested in event