Researchers have developed EMA-FS, a new method to accelerate the training of Gradient Boosted Decision Trees (GBDTs) like LightGBM. This technique optimizes the histogram construction process, which typically consumes a significant portion of training time. EMA-FS intelligently screens features based on their historical predictive utility, retaining high-gain features while discarding less useful ones, unlike random subsampling methods. The approach has demonstrated substantial speedups on various datasets, with potential improvements in AUC scores. AI
IMPACT This optimization could significantly reduce training times for GBDT models, enabling faster iteration and deployment in applications like fraud detection and click-through prediction.
RANK_REASON The cluster contains an academic paper detailing a new algorithm for accelerating machine learning model training.
Read on Hugging Face Daily Papers →
AI-generated summary · Google Gemini · from 2 sources. How we write summaries →
