1 paper · 1 filter
Tianyi Yin, Jingwei Wang, Yunlong Ma +6
Encoding time series into tokens and using language models for processing has been shown to substantially augment the models' ability to generalize to unseen tasks. However, existi…