Exploring the efficacy of transfer learning in mining image-based software artifacts

BackgroundTransfer learning allows us to train deep architectures requiring a large number of learned parameters, even if the amount of available data is limited, by leveraging existing models previously trained for another task. In previous attempts to classify image-based software artifacts in the absence of big data, it was noted that standard off-the-shelf deep architectures such as VGG could not be utilized due to their large parameter space and therefore had to be replaced by customized architectures with fewer layers. This proves to be challenging to empirical software engineers who would like to make use of existing architectures without the need for customization.FindingsHere we explore the applicability of transfer learning utilizing models pre-trained on non-software engineering data applied to the problem of classifying software unified modeling language (UML) diagrams. Our experimental results show training reacts positively to transfer learning as related to sample size, even though the pre-trained model was not exposed to training instances from the software domain. We contrast the transferred network with other networks to show its advantage on different sized training sets, which indicates that transfer learning is equally effective to custom deep architectures in respect to classification accuracy when large amounts of training data is not available.ConclusionOur findings suggest that transfer learning, even when based on models that do not contain software engineering artifacts, can provide a pathway for using off-the-shelf deep architectures without customization. This provides an alternative to practitioners who want to apply deep learning to image-based classification but do not have the expertise or comfort to define their own network architectures.