거대 다변수 함수 최적화 문제 해결을 위한 L-BFGS-B 알고리즘의 연산 가속화 연구

초록

The L-BFGS-B algorithm is widely recognized for its efficiency in multi-variable, bound-constrained optimization workloads, but high-dimensional problems remain computationally demanding because of their heavy time cost. Here we address this challenge through parallel computing, using CUDA for GPU-based acceleration and OpenMP for acceleration on multi-core CPUs. The experiments show that our approach substantially reduces the time needed to obtain an optimal parameter set for large-scale multivariate functions, reaching a speedup of up to ∼18,000x over computation on a single CPU core. These gains extend the capability of the L-BFGS-B algorithm to the complex high-dimensional optimization tasks that arise across a wide range of scientific and engineering applications.

키워드