GPU-Accelerated Record Linkage

• Bit-parallel Jaro similarity and GPU-accelerated record-linkage algorithms that process millions of records and billions of candidate comparisons, with more than 100× end-to-end speedup over the CPU baseline.
• GPU string-similarity kernels achieving roughly 381×–702× speedup over CPU implementations in isolated similarity-computation experiments.
• Techniques include workload scheduling, persistent GPU kernels, and shared-memory optimization.

Manuscript submitted to the 2026 IEEE International Conference on Big Data.