Method evidence record

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

Source record

Citations copied verbatim from the method’s source record. No claim-level verification is inferred from them.

MICN (Multi-scale Isometric Convolution Network)

Taxonomic method record · 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

Open full method

Curated claims

Claims persisted in the evidence ledger, each with its own assessment.

No curated claims yet

This view does not invent a claim assessment when the ledger has none.

Related methods

Generated from the method graph and shown as machine-suggested relations — no evidence claim is inferred.

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