Introduction: The problem of multi-category classification attracts a lot of attention in the field of machine learning. Traditional solutions tend to change it into multiple binary problems. The corresponding strategies include decision trees, neural networks, and so on. This article is included in PRICAI 2016 Conference and proposes a maximally spaced binary tree error correction output code. The corresponding coding matrix is ​​more stable and easy to identify.
Title: Maximum Spacing Binary Tree Error Correction Output Code
Abstract: Encoding is the most important step in error correction output codes (ECOCs). Traditional coding strategies are usually data-independent. Recently, some tree coding algorithms have been proposed to take advantage of interactive information to predict inter-class separability. Binary trees are hierarchically partitioned from top to bottom to obtain an encoding matrix. However, these standards are usually calculated by a non-parametric method. Generally speaking, they require a large number of samples and are more likely to produce unstable results. In this paper, we propose a novel encoding algorithm that uses a maximum spacing between classes as a reference to construct a bottom-up binary tree based on maximum spacing. Therefore, the corresponding coding matrix is ​​more stable and easy to identify in the following classification. Experimental results show that our algorithm performs better than some of the most advanced encoding algorithms in ECOC.
Keywords: multi-class classification; maximum pitch binary tree; error correction output coding
The first author introduction:
Fa Zheng
Department of Computer Science, Southeast University, Key Laboratory of Computer Network and Information Integration, Ministry of Education.
Via PRICAI 2016
Original paper download
Lei Feng Network Press: This article by Lei Feng network (search "Lei Feng network" public number concern) exclusive compilation, without permission prohibited reprint!
