CodeJeNN:用于物理应用的简单C++神经网络生成器
CodeJeNN: A simple C++ neural network generator for physics applications
浏览论文内容
中文总结 AI 辅助
研究针对物理应用中机器学习集成Python库产生性能瓶颈的问题,提出CodeJeNN通过从训练的Keras模型自动生成自包含C++代码实现推理,消除外部依赖,经测试展示了加速效果。
中文摘要 AI 辅助
机器学习在物理应用数值方法中实现了加速,但将基于Python的库集成到高性能C++求解器中会产生性能瓶颈。我们展示了CodeJeNN,它通过从训练的Keras模型自动生成自包含的C++代码进行推理来弥合这一差距。这通过最小化内联函数消除了外部依赖,允许无缝集成到现有框架中。我们描述了Keras到C++的工作流程、支持的架构和局限性。通过在急切和JIT模式下针对Keras的推理基准测试以及模拟氢 - 空气混合层中粘度的CFD测试案例展示了CodeJeNN,在不牺牲准确性的情况下实现了加速。
英文摘要
Machine learning has shown speedups for numerical methods in physics applications, but integrating Python-based libraries into high-performance C++ solvers creates performance bottlenecks. We present CodeJeNN, which bridges this gap by auto-generating self-contained C++ code from trained Keras models for inference. This eliminates external dependencies through minimal inlined functions, allowing seamless integration into existing frameworks. We describe the Keras-to-C++ workflow, supported architectures, and limitations. CodeJeNN is demonstrated through inference benchmarks against Keras in eager and JIT modes and a CFD test case modeling viscosity in a hydrogen-air mixing layer, showing speedups without sacrificing accuracy.
发表机构
- San Diego State University(圣地亚哥州立大学)
- Aerospace Engineering Department(航空航天工程系)
- U.S. Naval Research Laboratory(美国海军研究实验室)
- Laboratories of Computational Physics and Fluid Dynamics(计算物理学与流体动力学实验室)
机构由 AI 辅助整理,请以论文原文为准。