Comparer des méthodes

	MICN	SCINet
Domaine	Apprentissage profond	Apprentissage profond
Famille	Machine learning	Machine learning
Année d'origine	2023	2022
Auteur d'origine	Huiqiang Wang et al.	Minhao Liu et al.
Type	CNN-based time-series forecasting architecture	Hierarchical convolutional time-series forecasting network
Source fondatrice	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.	Liu, M., Zeng, A., Chen, M., Xu, Z., Lai, Q., Ma, L., & Xu, Q. (2022). SCINet: Time series modeling and forecasting with sample convolution and interaction. NeurIPS.
Alias	Multi-scale Isometric Convolution Network, Multi-scale Local and Global Context Model, MICN Forecaster, Çok Ölçekli İzometrik Evrişim Ağı	Sample Convolution and Interaction Network, SCI-Net, Temporal Downsampling Convolution Network, Örneklem Evrişim ve Etkileşim Ağı
Apparentées	2	2
Résumé	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.	SCINet is a deep learning architecture for multi-step time-series forecasting introduced by Liu et al. at NeurIPS 2022. Its core idea is a recursive binary-tree structure of SCI-Blocks, each of which splits an input sequence into odd- and even-indexed sub-sequences, applies convolutional filters to model cross-subsequence interactions, and then merges the learned representations. This hierarchical downsampling strategy enables the network to capture temporal dependencies at multiple resolutions simultaneously.