Many cognitive techniques, such as recognition and categorization are assumed to have need of establishing similarities between perceptual or conceptual representations. Basically when facing situations similar to what we have encountered before. This phenomenon signalled the development of many mathematical models of similarity.
Metric learning is one amongst those models that perform the task of devising similarities over objects. It has a plethora of applications in fields like information retrieval and recommendation system. Also, many machine learning approaches rely on some metric. This includes unsupervised techniques such as clustering, supervised procedures like KNN classification and semi-supervised modes as well. Metric learning has been involved as a preprocessing step for many of these approaches.
Henceforth learning a neat distance metric is a crucial task for forming similarities. LMNN & BoostMetric are among the supervised distance metric learning techniques which excel in this task. Accordingly, the fundamental objective of this project is to code novel implementations of LMNN & BoostMetric, with the purpose of achieving better benchmarking over kNN classification.