A collection of papers and related materials dedicated to partial label learning.
Partial Label Learning is an emerging weakly-supervised learning framework where each training example is associated with multiple candidate labels among which only one label is valid. The key assumption of PLL is that the ground-truth label is concealed in the candidate label set but remains unknown. The task of partial label learning is to learn a multi-class classification model from the partial label training examples, which can assign proper class label for the unseen instance in prediction phase. The task of learning from examples with candidate label sets naturally arises under many real-world scenarios, such as web mining, multimedia content analysis, ecoinformatics, natural language processing, etc.
- A brief introduction to weakly supervised learning (National Science Review 2018) - This article is a classic survey on weakly-supervised learning.
- Disambiguation-free partial label learning (SCIENCE CHINA Information Sciences 2019) - This article is an up-to-date survey on partial label learning.
- Learning with multiple labels (NIPS'03) - A maximum likelihood approach with expectation maximization (PL-EM).
- Classification with partial labels (KDD'08) - PL-SVM which maximize the margin between the maximum output of candidate labels and maximum output of non-candidate labels.
- A conditional multinomial mixture model for superset label learning (NIPS'12) - Logistic StickBreaking Conditional Multinomial Model (LSB-CMM).
- Maximum margin partial label learning (ACML'15) - M3PL which maximize the margin between the maximum output of candidate labels and the maximum output of any other labels.
- Maximum margin partial label learning (Machine Learning 2017) - A extended version of M3PL in ACML'15.
- Leveraging latent label distributions for partial Label Learning (IJCAI'18) - partial label learning with LAtent Label distributiOns (LALO) which regresses the label confidence matrix with local consistency.
- Partial label learning with self-guided retraining (AAAI'19) - Self-gUided REtraining (SURE) which regresses the label confidence matrix with infinite norm of label confidence matrix.
- Partial Label Learning by Semantic Difference Maximization (IJCAI'19) - Semantic DIfference Maximization (SDIM) which regresses the label confidence matrix with semantic maximization.
- Learning from ambiguously labeled examples (Intelligent Data Analysis 2006) PL-KNN.
- Learning from ambiguously labeled images (CVPR'09) A shorter version of CLPL in JMLR 2011.
- Learning from partial labels (JMLR 2011) - Convex Learning from Partial Labels (CLPL) which distinguish the average output of candidate labels from that of non-candidate labels.
- A regularization approach for instance-based superset label learning (TCYB 2018)
- Learning from candidate labeling sets (NIPS'10) MaximumMargin Set learning (MMS)
- Ambiguously labeled learning using dictionaries (TIFS 2014)
- Learning from Ambiguously Labeled Face Images (TPAMI 2018)
- Partial Label Learning with Batch Label Correction (AAAI'20)
An MLP-based approach for batch label predction and uses tricks of mixUp and knowledge distillation