Description: NMLP is a package for non-metric label propagation. Label propagation is a popular scheme for exploiting unlabeled data when a good graph can be constructed by considering the pairwise distances between training examples including both labeled and unlabeled data. When the original distance is non-metric, e.g, the triangular inequality axiom does not hold (this is often encountered in applications involving perceptual distances, since human perception does not necessarily satisfy the mathematical distance metric axioms), however, a direct application of current label propagation methods will lead to conflicts and mistakes. The NMLP provides an approach to label propagation in such cases.
Reference: Y. Zhang and Z.-H. Zhou. Non-metric label propagation. In: Proceedings of the 21st International Joint Conference on Artificial Intelligence (IJCAI'09), Pasadena, CA, 2009, pp.1357-1362.
ATTN: This package is free for academic usage. You can run it at your own risk. For other purposes, please contact Prof. Zhi-Hua Zhou (zhouzh@nju.edu.cn).
Requirement: The package was developed with MATLAB (2008b). The Java routines can be called in MATLAB environment.
ATTN2: This package was developed by Mr. Yin Zhang (zhangyin@lamda.nju.edu.cn). For any problem concerning the code, please feel free to contact Mr. Zhang.
Download:
code (4Kb)