HPAC-ML: A Programming Model for Embedding ML Surrogates in Scientific Applications

The end of Dennard scaling and the slowdown of Moore's Law led to heterogeneous architectures benefiting machine learning (ML) algorithms. These hardware advancements and the development of intuitive domain-specific languages have made ML more accessible, democratizing innovation. ML models surpass traditional approximation limits, broadening opportunities and evolving from statistical to complex function modeling. Consequently, scientific applications leverage ML models for enhanced execution speeds. However, integrating ML models remains manual and complex, slowing the adoption of ML as an approximation technique in modern applications. We propose an easy-to-use directive-based programming model that enables developers to describe the use of ML models in scientific applications. The runtime support, as instructed by the programming model, performs data assimilation using the original algorithm and can replace the algorithm with model inference. Our evaluation across five benchmarks, testing over 5000 ML models, shows up to 83.6x speed improvements with minimal accuracy loss (as low as 0.01 RMSE).