Predicting Gene Ontology Annotations from CAFA Using Distance Machine Learning and Transfer Metric Learning

Summary

Distance metric learning (DML) aims to determine how best to expose the information underlying the relationship. It is essential to many machine learning methods, pattern recognition, and data mining techniques, and to function well, these algorithms typically need a lot of label information, such as class labels or pair/triplet constraints. Nevertheless, the label data might need to be more adequate in real‐world applications since labelling is expensive, and DML might not work. Transfer metric learning (TML) can help reduce this problem for DML in the area of interest by utilizing knowledge and information from other relevant domains. Transfer metric learning has yet to progress while reaching a certain degree of development in several areas, including handling massive data, complex data, theoretical comprehension, and extreme circumstances. We offer a comprehensive analysis of the TML literature in this survey. Precisely, we classify TML based on configurations and metric transfer techniques, including distribution approximation, subspace estimation, distance approximation, and direct metric approximation. There will be a summary and enlightening discussion of the several TML techniques and their uses. Lastly, we list a few difficulties and offer some potential future paths.