Combining Supervised Learning and Reinforcement Learning for Multi-Label Classification Tasks with Partial Labels
This addresses annotation challenges in multi-label tasks for domains like document analysis and image classification, though it appears incremental as it builds on existing RL and supervised learning methods.
The paper tackles the multi-label positive-unlabelled learning problem, where only a subset of positive classes is annotated, by proposing MLPAC, a framework that combines reinforcement learning and supervised learning, achieving competitive performance across tasks like document-level relation extraction and multi-label image classification.
Traditional supervised learning heavily relies on human-annotated datasets, especially in data-hungry neural approaches. However, various tasks, especially multi-label tasks like document-level relation extraction, pose challenges in fully manual annotation due to the specific domain knowledge and large class sets. Therefore, we address the multi-label positive-unlabelled learning (MLPUL) problem, where only a subset of positive classes is annotated. We propose Mixture Learner for Partially Annotated Classification (MLPAC), an RL-based framework combining the exploration ability of reinforcement learning and the exploitation ability of supervised learning. Experimental results across various tasks, including document-level relation extraction, multi-label image classification, and binary PU learning, demonstrate the generalization and effectiveness of our framework.