New EMA-FS algorithm speeds up GBDT training by screening features

Researchers have developed EMA-FS, a novel algorithm-level optimization designed to accelerate the training of Gradient Boosted Decision Trees (GBDTs), such as LightGBM. This method addresses the significant time spent on per-feature histogram construction by intelligently screening features based on their historical predictive utility. EMA-FS maintains an exponential moving average of feature split gains and prioritizes histogram construction for the top-performing features, offering a more informed approach than random feature subsampling. AI

IMPACT This optimization could lead to faster model development cycles for applications relying on GBDT models.

RANK_REASON The cluster contains an academic paper detailing a new algorithm for machine learning model training. [lever_c_demoted from research: ic=1 ai=1.0]

Read on arXiv cs.LG →

• EMA-FS
• LightGBM

• paper
• infra

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