arXiv 2310.10688
A decoder-only foundation model for time-series forecasting
By Abhimanyu Das, Weihao Kong, et al.
Published 2023-10-14
Wiki summary
Explore the paper's summary, context, and related research on Papiers.
Motivated by recent advances in large language models for Natural Language Processing (NLP), we design a time-series foundation model for forecasting whose out-of-the-box zero-shot performance on a variety of public datasets comes close to the accuracy of state-of-the-art supervised forecasting models for each individual dataset. Our model is based on pretraining a patched-decoder style attention model on a large ti…