Efficient Tensor Representation for Deep Learning with TensorLy and PyTorch

Explore efficient tensor representations for deep learning in this 31-minute conference talk by Dr. Jean Kossaifi from Nvidia Corporation. Discover how preserving and leveraging multi-dimensional data structure using tensor methods can lead to better representations and improved learning, especially for spatiotemporal and structured data like MRI. Learn about tensor methods for deep learning that enhance performance, speed, model compression, and robustness. Gain insights into practical implementation using PyTorch and TensorLy-Torch, with examples of improving ResNet models for video-based classification on the Kinetics dataset and large-scale image classification on ImageNet. Presented at the Institute for Pure & Applied Mathematics (IPAM) workshop on Tensor Methods and Emerging Applications to the Physical and Data Sciences 2021, this talk offers valuable knowledge for researchers and practitioners in the field of deep learning and tensor methods.

Jean Kossaifi: "Efficient Tensor Representation for Deep Learning with TensorLy and PyTorch"