PulseAugur
LIVE 07:45:10
research · [2 sources] ·
0
research

New MELO method hedges memory horizons for non-stationary prediction

Researchers have developed MELO, a novel model-agnostic method for online prediction that hedges across different adaptation scales. MELO wraps base predictors with exponentially weighted least-squares adaptation experts and aggregates their forecasts using a parameter-free online aggregation rule. In evaluations on French electricity load forecasting during the COVID-19 lockdown, MELO significantly reduced RMSE compared to baseline methods and a reference model that used external policy covariates. AI

Summary written by gemini-2.5-flash-lite from 2 sources. How we write summaries →

IMPACT Introduces a new method for handling non-stationary data in prediction tasks, potentially improving forecasting accuracy in dynamic environments.

RANK_REASON This is a research paper detailing a new prediction method.

Read on arXiv stat.ML →

COVERAGE [2]

  1. arXiv cs.LG TIER_1 · Yutong Wang, Yannig Goude, Qiwei Yao ·

    Hedging Memory Horizons for Non-Stationary Prediction via Online Aggregation

    arXiv:2605.06541v1 Announce Type: new Abstract: We study online prediction under distribution shift, where inputs arrive chronologically and outcomes are revealed only after prediction. In this setting, predictors must remain stable in quiet regimes yet adapt when regimes shift, …

  2. arXiv stat.ML TIER_1 · Qiwei Yao ·

    Hedging Memory Horizons for Non-Stationary Prediction via Online Aggregation

    We study online prediction under distribution shift, where inputs arrive chronologically and outcomes are revealed only after prediction. In this setting, predictors must remain stable in quiet regimes yet adapt when regimes shift, and the right adaptation memory is unknown in ad…