Received: June 15, 2022. Revised: July 22, 2022. 83

learning method. • We compared the symptoms, causes, and repair patterns we had found. We discover that TensorFlow contains type confusions depending on the comparison, which the earlier studies had not mentioned. • Additionally, we discover that TensorFlow suffers dimension mismatches just like deep … learning applications. Some of the issues they observed aren't found in deep learning libraries or are unusual. They discover problems like improper model parameters and structure inefficiency in TensorFlow applications, for example. Crashing accounts for 92% of TensorFlow difficulties, with performance bugs accounting for the remaining 8%. … and the causes are more important than the symptoms. (2) There are several similarities between regular software defects and TensorFlow bugs; (3) Inconsistent defects are widespread in other supporting components in contrast; API implementations with inadequate data formatting (dimension and type) are prone to problems. They