Metodebevisregistrering

MICN

MICN (Multi-scale Isometric Convolution Network) is a convolutional neural network architecture for long-term time-series forecasting introduced by Huiqiang Wang and colleagues at ICLR 2023. Its central idea is to capture both local temporal patterns and global seasonal dependencies simultaneously through multi-scale isometric convolutions combined with a merge attention mechanism, enabling efficient and expressive modeling of complex temporal dynamics without the quadratic cost of full self-attention.

Sources recorded, not reviewed

Kilderegistrering

Citater kopieret ordret fra metodens kilderegistrering. Ingen påstandsniveauverifikation er udledt heraf.

MICN (Multi-scale Isometric Convolution Network)

Taksonomisk metoderegistrering · ml-model / deep-learning

Wang, H., Peng, J., Huang, F., Wang, J., Chen, J., & Xiao, Y. (2023). MICN: Multi-scale local and global context modeling for long-term series forecasting. ICLR. · URL

Åbn fuld metode

Kuraterede påstande

Påstande gemt i bevis-loggen, hver med sin egen vurdering.

Ingen kuraterede påstande endnu

Denne visning opfinder ikke en påstandsvurdering, når loggen ingen har.

Relaterede metoder

Genereret fra metodegrafen og vist som maskinelt foreslåede relationer — ingen bevispåstand er udledt.

Taxonomic bucketSCINetmachine-suggested · Relational suggestion, not evidence.Taxonomic bucketTimesNetmachine-suggested · Relational suggestion, not evidence.