Ord2Seq: Relating to Ordinal Regression as Label Sequence Prediction
Authors: Jinhong Wang, Yi Cheng, Jintai Chen, Tingting Chen, Danny Chen, Jian Wu
Summary: Ordinal regression refers to classifying object cases into ordinal classes. It has been broadly studied in lots of situations, resembling medical illness grading, film ranking, and many others. Identified strategies targeted solely on studying inter-class ordinal relationships, however nonetheless incur limitations in distinguishing adjoining classes to date. On this paper, we suggest a easy sequence prediction framework for ordinal regression referred to as Ord2Seq, which, for the primary time, transforms every ordinal class label right into a particular label sequence and thus regards an ordinal regression activity as a sequence prediction course of. On this means, we decompose an ordinal regression activity right into a sequence of recursive binary classification steps, in order to subtly distinguish adjoining classes. Complete experiments present the effectiveness of distinguishing adjoining classes for efficiency enchancment and our new strategy exceeds state-of-the-art performances in 4 totally different situations. Codes can be found at https://github.com/wjh892521292/Ord2Seq.
