Semi-supervised Metric Learning
Semi-supervised metric learning learns a task-adapted distance function by combining a small set of labeled pairwise constraints — must-link and cannot-link pairs — with the geometric structure of a much larger pool of unlabeled data. The result is a Mahalanobis-style or kernel-based distance that reflects both supervision and data topology, improving downstream tasks such as nearest-neighbor classification and clustering.
Kilderegistrering
Citater kopieret ordret fra metodens kilderegistrering. Ingen påstandsniveauverifikation er udledt heraf.
- Yeung, D.-Y., & Chang, H. (2007). A kernel approach for semi-supervised metric learning. IEEE Transactions on Neural Networks, 18(1), 141–149. · DOI 10.1109/TNN.2006.883723
- Davis, J. V., & Dhillon, I. S. (2008). Structured metric learning for high dimensional problems. Proceedings of the 14th ACM SIGKDD International Conference on Knowledge Discovery and Data Mining, 195–203. · DOI 10.1145/1401890.1401918
Kuraterede påstande
Påstande gemt i bevis-loggen, hver med sin egen vurdering.
Denne visning opfinder ikke en påstandsvurdering, når loggen ingen har.
Relaterede metoder
Genereret fra metodegrafen og vist som maskinelt foreslåede relationer — ingen bevispåstand er udledt.