A Statistical Learning Model with Deep Learning Characteristics
Lei Liao, Zhiqiu Huang, Wengjie Wang · 2021
Although machine learning has achieved great success in many fields, the lack of interpretability and excessive computational amount and poor robustness severely limits its wide application in real-world tasks, especially security-sensitive tasks. But the current deep learning research is still far from truly solving these problems. In order to overcome these problems encountered by the deep learning model, this article does not intend to modify the deep learning model itself, but design a new machine learning model to avoid various problems of deep learning. We proposes a new statistical learning model that learn from the characteristics of deep learning. Then we evaluate these models on two datasets and found that the new model is significantly better than the deep learning model in terms of computational complexity, robustness and interpretability.