English (United Kingdom)

https://curated-unify.zendy.io/wp-json/zendy-region/v1/featured_content/oa?rat=en

https://curated-unify.zendy.io/wp-json/zendy-region/v1/highlighted_journal/

Global: Zendy Plus annual

Presents the access of premium content as premium feature

Premium Content

Presents the keyphrase highlighting as premium feature

Keyphrase Highlighting

Presents the summarisation as premium feature

Summarisation

Insights

Presents the pdf analysis as premium feature

PDF Analysis

Presents the zaia usage as premium feature

ZAIA

Global: Zendy Plus monthly

Global: Zendy Tools annual

Global: Zendy Tools monthly

Global: Zendy Free Plan

As Transformer-based models have achieved impressive performance on varioustime series tasks, Long-Term Series Forecasting (LTSF) tasks have also receivedextensive attention in recent years. However, due to the inherent computationalcomplexity and long sequences demanding of Transformer-based methods, itsapplication on LTSF tasks still has two major issues that need to be furtherinvestigated: 1) Whether the sparse attention mechanism designed by thesemethods actually reduce the running time on real devices; 2) Whether thesemodels need extra long input sequences to guarantee their performance? Theanswers given in this paper are negative. Therefore, to better copy with thesetwo issues, we design a lightweight Period-Attention mechanism (Periodformer),which renovates the aggregation of long-term subseries via explicit periodicityand short-term subseries via built-in proximity. Meanwhile, a gating mechanismis embedded into Periodformer to regulate the influence of the attention moduleon the prediction results. Furthermore, to take full advantage of GPUs for fasthyperparameter optimization (e.g., finding the suitable input length), aMulti-GPU Asynchronous parallel algorithm based on Bayesian Optimization (MABO)is presented. MABO allocates a process to each GPU via a queue mechanism, andthen creates multiple trials at a time for asynchronous parallel search, whichgreatly reduces the search time. Compared with the state-of-the-art methods,the prediction error of Periodformer reduced by 13% and 26% for multivariateand univariate forecasting, respectively. In addition, MABO reduces the averagesearch time by 46% while finding better hyperparameters. As a conclusion, thispaper indicates that LTSF may not need complex attention and extra long inputsequences. The code has been open sourced on Github.

Does Long-Term Series Forecasting Need Complex Attention and Extra Long Inputs?