Please use this identifier to cite or link to this item:
|Scopus||Web of Science®||Altmetric|
|Title:||A hybrid loss for multiclass and structured prediction|
Van Den Hengel, A.
|Citation:||IEEE Transactions on Pattern Analysis and Machine Intelligence, 2015; 37(1):2-12|
|Qinfeng Shi, Mark Reid, Tiberio Caetano, Anton van den Hengel, and Zhenhua Wang|
|Abstract:||We propose a novel hybrid loss for multiclass and structured prediction problems that is a convex combination of a log loss for Conditional Random Fields (CRFs) and a multiclass hinge loss for Support Vector Machines (SVMs). We provide a sufficient condition for when the hybrid loss is Fisher consistent for classification. This condition depends on a measure of dominance between labels-specifically, the gap between the probabilities of the best label and the second best label. We also prove Fisher consistency is necessary for parametric consistency when learning models such as CRFs. We demonstrate empirically that the hybrid loss typically performs least as well as-and often better than-both of its constituent losses on a variety of tasks, such as human action recognition. In doing so we also provide an empirical comparison of the efficacy of probabilistic and margin based approaches to multiclass and structured prediction.|
|Keywords:||Humans; Models, Statistical; Human Activities; Artificial Intelligence; Image Processing, Computer-Assisted; Video Recording; Pattern Recognition, Automated; Support Vector Machine|
|Rights:||© 2014 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission.|
|Appears in Collections:||Computer Science publications|
Files in This Item:
|RA_hdl_95549.pdf||Restricted Access||889.85 kB||Adobe PDF||View/Open|
Items in DSpace are protected by copyright, with all rights reserved, unless otherwise indicated.