CASA: CNN Autoencoder-based Score Attention for Efficient Multivariate Long-term Time-series Forecasting

Minhyuk Lee, HyeKyung Yoon, MyungJoo Kang · 2025

Read CASA: CNN Autoencoder-based Score Attention for Efficient Multivariate Long-term Time-series Forecasting by Minhyuk Lee, HyeKyung Yoon, MyungJoo Kang (2025) 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-2505-02011.txt.

Description

Minhyuk Lee, HyeKyung Yoon, and MyungJoo Kang present, in a 2025 paper, an add-on that makes Transformer forecasts of long time series cheaper and sharper. Long-range forecasts feed tasks like weather and traffic calls. Past add-ons still cost much compute and miss links across the parts of a series. Their part, CASA, runs the series through a small CNN autoencoder first. A score-based attention then weights what that squeeze kept. The part drops into many Transformer forms with no change to their core.

On eight real data sets, CASA cut compute use by up to 77.7 percent. It sped runs by 44 percent as well. Scores came first in 87.5 percent of the gauges used, a state-of-the-art mark for this task class.

Catalog record

CREATORMinhyuk Lee, HyeKyung Yoon, MyungJoo Kang
YEAR2025
COLLECTIONPAPERS
FORMATTXT
FILE SIZE41 KB
VERSION1.0
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2505-02011

Browse the full library · About the Tower of Babel