Accelerating XGBoost on ARM CPUs with Scalable Vector Extension for High-Performance Data Science
Tuesday, June 10, 2025 3:00 PM to Thursday, June 12, 2025 4:00 PM · 2 days 1 hr. (Europe/Berlin)
Foyer D-G - 2nd floor
Project Poster
ML Systems and ToolsNumerical LibrariesOptimizing for Energy and Performance
Information
Poster is on display.
Decision trees are a cornerstone of many machine learning algorithms, offering
interpretable & robust models for structured data. XGBoost (eXtreme Gradient
Boosting)[1] uses an ensemble of decision trees to deliver high performance in
gradient boosting.
In this work, we leverage ARM Scalable Vector Extension (SVE)[2], which is a
vector extension for Armv-8A that supports variable length vectors from 128 to
2048 bits. By utilizing SVE’s vectorization capabilities we accelerate XGBoost’s
training pipeline by optimizing the histogram update function - a key step in
constructing decision trees.
The results of our experiments on Higgs Boson dataset show a 2x speed-up in
training time compared to the non-SVE optimized code, with same accuracy on
ARM architectures.
Contributors:
Decision trees are a cornerstone of many machine learning algorithms, offering
interpretable & robust models for structured data. XGBoost (eXtreme Gradient
Boosting)[1] uses an ensemble of decision trees to deliver high performance in
gradient boosting.
In this work, we leverage ARM Scalable Vector Extension (SVE)[2], which is a
vector extension for Armv-8A that supports variable length vectors from 128 to
2048 bits. By utilizing SVE’s vectorization capabilities we accelerate XGBoost’s
training pipeline by optimizing the histogram update function - a key step in
constructing decision trees.
The results of our experiments on Higgs Boson dataset show a 2x speed-up in
training time compared to the non-SVE optimized code, with same accuracy on
ARM architectures.
Contributors:
Format
On DemandOn Site
