.png){kind=icon}
.png){kind=icon}
Return
A General-Purpose K-Nearest Neighbor Method with an Efficient Pruning Strategy for GPUs
.png){kind=icon}
OA
AI
J
F
DOI：10.1016/j.jpdc.2025.105187.png)
.png){kind=icon}
.png){kind=icon}
Share.png){kind=icon}
Save
Abstract
En 中文
• The proposed K-nearest neighbor method switches between the indexing and exhaustive approaches to improve pruning efficiency with varying data dimensionality. • Our method exploits the GPU architecture and enables fine-grained parallelism. • The candidate references in GPU shared memory are updated rapidly with parallel bitonic operations. • Experiments against seven baseline methods, including Meta’s Faiss library, on large-scale datasets with up to 10 million references and 960 dimensions are presented. • Our method achieves a 15.9 times speedup with L2 distances and a 36.7 times speedup with angular distances.
Keywords:
GPU
k-nearest neighbor
parallel computing
AI Summary
Key information extracted from the uploaded paper, including a brief overview, abstract, background, key highlights, visual analysis, and future outlook.
Journal
IF:
4
Papers:
3.8K
Citations:
4.8K
Organization
Cited Papers