ProPML: Probability Partial Multi-label Learning
This addresses the challenge of weakly supervised learning with noisy candidate labels, which is incremental as it builds on existing PML methods by improving robustness to noise.
The paper tackles the problem of Partial Multi-label Learning (PML), where training instances have candidate labels with some being true, by introducing ProPML, a probabilistic approach that extends binary cross entropy to PML without requiring suboptimal disambiguation, and it outperforms existing methods, particularly under high noise conditions.
Partial Multi-label Learning (PML) is a type of weakly supervised learning where each training instance corresponds to a set of candidate labels, among which only some are true. In this paper, we introduce \our{}, a novel probabilistic approach to this problem that extends the binary cross entropy to the PML setup. In contrast to existing methods, it does not require suboptimal disambiguation and, as such, can be applied to any deep architecture. Furthermore, experiments conducted on artificial and real-world datasets indicate that \our{} outperforms existing approaches, especially for high noise in a candidate set.