New method improves autoregressive forecasting by reliability-weighting gradients

Researchers have introduced Internal Dual-Wiener routing (Internal-DW), a novel method for improving autoregressive forecasting by addressing the unreliability of distant gradients in backpropagation through time (BPTT). This technique selectively weights internal gradient routes, balancing the preservation of predictable learning signals against the suppression of unpredictable noise. Experiments on history-dominated testbeds show Internal-DW reduces forecast error by up to 13.8% compared to standard BPTT and outperforms other regularization methods. AI

IMPACT Introduces a technique to enhance the reliability of gradient updates in long-horizon forecasting models.

RANK_REASON Academic paper detailing a new method for improving forecasting models. [lever_c_demoted from research: ic=1 ai=1.0]

Read on arXiv cs.LG →

• arXiv
• autoregressive forecasting
• backpropagation through time
• BPTT
• Hugging Face
• Internal Dual-Wiener routing
• Internal-DW

• paper
• model release

AI-generated summary · Google Gemini · from 1 sources. How we write summaries →