Recently, AI technology provides information such as classification, recommendation, and prediction to
users and is being used for various purposes in a wide range of fields. However, AI technology has a
limitation in that it does not provide validity for the final results derived due to the existence of black
box characteristics. To solve this problem, many studies on the XAI (eXplainable AI) model are being
conducted to explain the transparency and validity of the process and judgment grounds judged by the
deep learning model. This XAI model improves reliability by providing users with explanation of why
these results were derived. However, in the case of the XAI model, there is a problem that resources
are consumed as the number of objects to be processed increases. For example, when detecting objects
of various scales using a representative Feature Pyramid Network structure applied to improve explainability, visualized characteristics for each layer must be extracted, which consumes a lot of computing
resources. To solve this problem, task scheduling shall be performed in consideration of the processing
process in a heterogeneous environment with energy efficiency characteristics.
To solve the above problem, this thesis considers the explanatory of the multi-scale feature map of
FPN and the scheduling considering the resource cost of visual explanatory map generation. First, we
proposed explanatory modeling based on the visualization map of the Scalable Feature Block and the
object detection believer of the learned model. Since the explanatory properties of FPN Blocks differ
depending on the characteristics of the target object, we modeled the explanatory properties as a linear
model for the block reflection ratio of FPNs and detection performance by object. In addition, we propose
an adaptive XAI task scheduling algorithm to minimize energy consumption of heterogeneous FPGAGPU resources when generating a Saliency Map. The proposed scheduling considers the distribution of
tasks in sub-work units to minimize processing costs while satisfying the explanability. We evaluated
the proposed algorithm through multiple object detection applications in heterogeneous FPGA-GPU
clusters. Compared with existing accelerator-based scheduling, the proposed algorithm showed that it
could save 48% of energy consumption at the expense of less than 20% of processing time.