TIME SERIES FORECASTING WITH VISION MODELS

Summary from the TLDR AI newsletter

VisionTS is a new approach that leverages the similarities between natural images and time series data to enhance forecasting by reformulating time series forecasting as an image reconstruction task. It uses a visual masked autoencoder (MAE) pre-trained on ImageNet to achieve impressive zero-shot performance.