•
Wen, Zhou, Zhang, Chen, Ma, Yan and Sun (IJCAI 2023; arXiv:2202.07125) survey Transformers for time series. Taxonomy: module tweaks (positional encodings, sparse/low-rank attention) vs architecture (hierarchical). Applications: forecasting, anomaly, classification. Their ETTm2 checks: longer inputs often hurt; 3–6 layers beat 48; seasonal-trend decomposition can cut MSE 50–80%. Map before you pick Informer vs PatchTST. Plain-language intro. Not investment advice.
Takeaway. This is the map, not a new net. Wen et al. sort time-series Transformers by what they changed (positional encoding, attention sparsity, hierarchy) and what they are used for (forecast, anomaly, classify). Their ETTm2 homework: longer look-backs often hurt; fat stacks do not help; a cheap seasonal-trend split can dominate the attention trick.
Transformers in Time Series: A Survey (Qingsong Wen, Tian Zhou, Chaoli Zhang, Weiqi Chen, Ziqing Ma, Junchi Yan, Liang Sun, 2022; IJCAI 2023; arXiv:2202.07125). Alibaba DAMO / Shanghai Jiao Tong.
Fit for learners: Taxonomy of TS transformers — cheat-sheet before picking Informer/Autoformer/PatchTST for markets.
This Hub note is a plain-language intro. Not investment advice.
NLP/CV surveys of Transformers exist. TS has extra headaches: seasonality, mixed frequencies, multivariate channels that are not “words.” Without a taxonomy you re-read Informer, Autoformer, FEDformer, Pyraformer as five unrelated GitHubs.
Read this before you spend a weekend on ProbSparse vs patches.
The survey’s useful punchline is not “Transformers win.” It is: encode timestamps, consider decomposition, do not assume longer context helps, keep the stack shallow. That is a one-page design checklist for an evening notebook.
Tiny evening exercise: implement two positional encodings on a length-96 vector — sin/cos with a few frequencies, vs a learned nn.Embedding(96, d). Add each to a dummy series and print cosine-similarity of nearby vs far steps. Learning only.
Landing: arXiv abs
Not investment advice. This Hub note is educational only. A survey of LTSF Transformers is not a ranking of trading systems.
Single course CTA: SoloQuant Academy’s Introduction to Python Programming—lists, dicts, and plotting two encodings is how you start the survey’s positional-encoding section in code.