 This paper proposes a deep learning framework to estimate the state of health of batteries without requiring any target labels. It uses a combination of multiple deep neural networks to achieve high accuracy in predicting the state of health of batteries. The authors tested their model on 65 commercial batteries from 5 different manufacturers and found that it was able to accurately estimate the state of health of 89.4% of the samples with an average absolute error of less than 3%. Furthermore, the maximum absolute error was less than 8.87%, demonstrating the potential of deep learning to replace costly and time-consuming degradation experiments. This article was authored by Jia Huan Lu, Rui Xiong, Jin Peng Tian, and others.