Recurrent Halting Chain for Early Multi-label Classification

Published in KDD, 2020

Early Multi-label Classification is the marriage of the Early Classification and Multi-label Classification problems. Early Classification is the goal of classifying time series while using as few timesteps as possible. Multi-label classification is the goal of predicting a set of labels for each instance. The problem of Early Multi-label Classification is to both predict each label in the label set as early as possible while concurrently modeling the dependencies between labels themselves. This setting is extremely important in settings such as diagnosis, where a person may have multiple ailments concurrently and their set of ailments often come in groups (for instance, if a person has diabetes, there is a larger chance that they have heart disease than bone spurs). We solve this problem using a novel combination of Classifier Chains for multi-label classification and Adaptive-Halting Policy Networks for early classification, resulting in our proposed Recurrent Halting Chain.

@inproceedings{hartvigsen2020recurrent,
  title={Recurrent Halting Chain for Early Multi-label Classification},
  author={Hartvigsen, Thomas and Sen, Cansu and Kong, Xiangnan and Rundensteiner, Elke},
  booktitle={Proceedings of the 26th ACM SIGKDD international conference on Knowledge discovery and data mining},
  pages={101--110},
  year={2020},
  organization={ACM}
}