Auto-tuning (AT) is one of promising technologies not only to establish exascale computing but also to solve complex tuning procedures. For example, accuracy tuning for computing results has different difficulty to tune execution speed. It requires multiple conditions to optimize it with respect to tuning on speed. To reduce cost of the tunning of accuracy, AT can provide reasonable solutions. On the other hand, AI technologies are widely-used even in facilities based on AT. As same as problem of explainable AI (XAI), explainable AT (XAT) will be one of important issues in very near future.
With respect to the above background, two topics are folded in this talk. First, we will explain a case of accuracy optimization by AT with a scientific computation from filed of weather-earth simulation. Second, we will demonstrate XAI for parameter tuning on a numerical library from accuracy assurance computations.