MLBBQ: “Are Transformers Effective for Time Series Forecasting?” by Joanne Wardell

“Are Transformers Effective for Time Series Forecasting?”. This was one of the “most Influential AAAI Papers” of 2023 based on citations. The authors introduce a set of embarrassingly simple one-layer linear models named LTSF-Linear for comparison to Transformers.
arxiv.org/abs/2205.13504
www.paperdigest.org/2023/09/m...

[@DunlopJ]

No mention of PatchTST Transformer from "A Time Series is Worth 64 Words: Long-term Forecasting with Transformers"?

[@mmattb]

Great discussion and a sharp group! Around 32:27 one of the commenters mentioned a paper analyzing noisy multivariate dynamical systems and that under the CLT they may appear linear. Are you able to link me to that result? I'd love to see it.

[@SergeyPlis]

here's an empirical look at the brain arxiv.org/abs/2012.12351
and theoretical papers, that I guess have inspired that investigation:
Spatial averaging: ieeexplore.ieee.org/document/9993260
Temporal averaging: ieeexplore.ieee.org/abstract/document/10155808
HTH

[@mmattb]

@@SergeyPlis those looks like good fun! Thanks a bunch Sergey.

[@pranavishnoi20]

So we can say this paper is Haox