Researchers have developed RATL, a novel method for robust multivariate time-series forecasting that leverages retrieved historical forecast residuals. Unlike traditional approaches that discard residuals, RATL stores them as a memory specific to a base forecasting model. At inference time, RATL retrieves and combines these residual trajectories based on the current context, using a set-aware router to select and integrate them. Experiments demonstrate that RATL significantly improves the performance of base forecasters, such as iTransformer, across various benchmarks by providing a plug-in paradigm for learned feedback correction. AI
IMPACT This method could enhance the accuracy and robustness of forecasting systems across various domains by leveraging historical error patterns.
RANK_REASON The cluster contains an academic paper detailing a new method for time-series forecasting. [lever_c_demoted from research: ic=1 ai=1.0]
AI-generated summary · Google Gemini · from 1 sources. How we write summaries →
