论文部分内容阅读
This study proposes a supervised learning method that does not rely on labels.We use variables associated with the label as indirect labels,and construct an indirect physics-constrained loss based on the physical mechanism to train the model.In the traini