Exploring AI for Auto-tuning through Sparse Matrix Image Information

In recent years, significant strides have been made in the field of artificial intelligence (AI), particularly exemplified by the emergence of large-scale language models (LLM). On the other hand, the domain of automatic performance tuning (AT) for numerical computations has been a longstanding focus. Remarkably, the prerequisites for AT functionality closely align with those of AI. In fact, the application of AI to AT is conceptually straightforward. This presentation will delve into an AT methodology designed to handle the unique data characteristic of numerical computations ? sparse matrices, by representing them as images. More specifically, we will show a case of AT functions constructed through AI learning, utilizing sparse matrix image data for sparse iterative methods in sparse matrix computations.