Partial Multi-Label Learning via Large Margin Nearest Neighbour Embeddings

To deal with ambiguities in partial multi-label learning (PML), existing popular PML research attempts to perform disambiguation by direct ground-truth label identification. However, these approaches can be easily misled noisy false-positive labels the iteration of updating model parameter and latent variables. When labeling information is ambiguous, we should depend more on underlying structure data, such as feature correlations, for partially labeled data. Moreover, large margin nearest neighbour (LMNN) a strategy that considers data classification. due ambiguity PML, traditional LMNN cannot used solve problem directly. In addition, embedding an effective technology decrease noise Inspried technology, propose novel paradigm called Partial Multi-label Learning via Large Margin Nearest Neighbour Embeddings (PML-LMNNE), which aims conduct projecting features into lower-dimension space reorganize simultaneously. An efficient algorithm designed implement proposed method convergence rate analyzed. present theoretical analysis generalization error bound PML-LMNNE, shows converges sum two times Bayes over when number instances goes infinity. Comprehensive experiments artificial real-world datasets demonstrate superiorities PML-LMNNE.