A Wave is Worth 100 Words: Investigating Cross-Domain Transferability in Time Series

Xiangkai Ma, Xiaobin Hong, Wenzhong Li, Sanglu LuState Key Laboratory for Novel Software Technology, Nanjing University, Nanjing, China · 2024

Read A Wave is Worth 100 Words: Investigating Cross-Domain Transferability in Time Series by Xiangkai Ma, Xiaobin Hong, Wenzhong Li, Sanglu LuState Key Laboratory for Novel Software Technology, Nanjing University, Nanjing, China (2024) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2412-00772.txt.

Introduction

Xiangkai Ma, Xiaobin Hong, Wenzhong Li, and Sanglu Lu ask whether what a model learns from one kind of time series can transfer to others. Forecasting and classifying series, from heartbeats to markets, usually needs labeled data, which is costly, while unlabeled series pile up unused. Their 2024 paper proposes wave quantization, a pretraining method that chops series into wave pieces a model can learn from across domains. The method, WQ4TS, can be combined with existing backbones. It targets the double difficulty of domains shifting and tasks differing.

Transfer learning remade language and vision; time series still waits for its version. A shared wave vocabulary is the authors' candidate bridge. Their title promises that a wave may carry the weight of many words. The work is tested on cross-domain, multi-task migration. If the bridge holds, every unlabeled archive becomes training capital.

Catalog record

CREATORXiangkai Ma, Xiaobin Hong, Wenzhong Li, Sanglu LuState Key Laboratory for Novel Software Technology, Nanjing University, Nanjing, China
YEAR2024
COLLECTIONPAPERS
FORMATTXT
FILE SIZE161 KB
VERSION1.0
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2412-00772

Browse the full library · About the Tower of Babel