An Uncertainty Network Working Mechanism Analysis Method based on Local Interpretable Model-agnostic Explanations
Zhonghao Cheng, Yinxiao Miao, Xiujian Zhang · 2022
In recent years, Deep learning has been widely used in different fields which learns the skills from dataset, however the parameter and decision logic is opaque for human. The trustworthy of deep learning model has attracted much attention of researchers, in that uncertainty is an essential part of trusted artificial intelligence. Therefore uncertainty analysis is an important part of the basic theory of deep learning that needs to be researched. The stochastic differential equation network is able to predict the uncertainty by the design of network and training, but it’s doubtful for the uncertainty quantification. To analyze the deeply work mechanism in uncertainty prediction, we propose the an uncertainty network working mechanism analysis method based on local interpretable model-agnostic explanations. The explainable artificial intelligence method is applied to expose the decision logic of uncertainty prediction. By the experiments we have proved that the uncertainty generated by the stochastic differential equation network is related to the performance of prediction, and they have a positive correlation.