Inferencing the best AI service using Deep Neural Networks
Atsushi Hoshino, Takumi Saito, Mizuki Oka · 2019
Many learned inference engines have been released as cloud-based AI services. However, because these engines are easy to use, accuracy and fairness are necessary. A previous study revealed that the accuracy of cloud-based AI services varies depending on the input. Thus, we propose to infer the best AI services by learning their different output results as training data. Our model, "AI for selecting the best AI," involves meta-learning; it learns the output of the cloud-based AI service as metadata. The results of our experiments to infer face attributes (i.e., age and gender) on a face image dataset crawled from Wikipedia showed that the accuracy of our system was 2.4% higher than that of existing face classification cloud-based AI services. Notably, results on inferring age, where training data for each service showed a significant difference in the tendency for accuracy, had 2.7% higher accuracy compared to existing cloud-based AI services.