SACQ：面向长周期预测的记忆条件精炼结构化解码方法

View PDF HTML (experimental)

Abstract:Long-term time series forecasting (LTSF) models predominantly employ patch-based encoders terminated by a flatten readout head that maps the entire encoded historical memory to all future steps through a single shared projection. This implicit coupling of future positions obscures position-specific historical-to-future alignment and amplifies sensitivity to corrupted inputs and extreme supervision noise. We present SACQ, a plug-in structured prediction head that replaces flatten readout while keeping the encoder unchanged. SACQ adopts a two-stage decoding pipeline: it first establishes a coarse patch-grid forecast scaffold, then refines each future position through cross-attention over historical memory and merges the attention-derived correction with the coarse scaffold via a learned per-patch gate. To stabilize optimization under long horizons and noisy labels, we further propose a batch-adaptive scaled log-cosh loss that automatically calibrates robustness to the current residual scale, suppressing outlier gradients while preserving MSE-like sensitivity for typical errors. SACQ attains top-tier test MSE/MAE across PatchTST, DLinear, and patch-Mamba backbones with only modest incremental overhead in parameters and latency. Under inference-time input corruption and training-set label-noise stress tests, SACQ substantially outperforms flatten readouts, with ablation studies validating each architectural component.

Comments:	10 pages, 7 figures
Subjects:	Machine Learning (cs.LG)
Cite as:	arXiv:2610.11170 [cs.LG]
	(or arXiv:2610.11170v1 [cs.LG] for this version)
	https://doi.org/10.48550/arXiv.2610.11170 arXiv-issued DOI via DataCite (pending registration)

Submission history

From: Guo Cheng [view email]
[v1] Thu, 8 Oct 2026 03:25:22 UTC (1,032 KB)