.png){kind=icon}
返回
Incremental and dynamic graph construction with application to image classification
.png){kind=icon}
PRE
AI
DOI：10.1016/j.eswa.2019.113117.png)
.png){kind=icon}
原文求助.png){kind=icon}
分享.png){kind=icon}
收藏
摘要
En 中文
In this paper, we propose a dynamic graph construction technique that inserts new samples into a previously constructed graph, which reduces the computational time and complexity of the classic batch construction schemes. The basic assumption of the proposed method is that by adding one sample into a graph, only its close nodes will be affected, hence, it is not necessary to update the whole graph but only the weights of close nodes. The proposed method has two steps of insertion and updating. In the insertion phase, the similarity between the new sample and the available data is calculated. The similarity vector is retrieved either from a distance function or a coding scheme. Then, in the update phase, by evaluating the similarity vector of the new sample, close nodes which will be affected are identified and the graph weights of these close (similar) nodes are updated. By adopting this scenario, in each insertion of a node (or a set of nodes), only the weights of very few nodes have to be updated. It is worthy to mention that since the proposed method does not invoke labels of the samples, it can be adopted by any unsupervised, semi-supervised or supervised technique. A set of extensive experiments for the task of classification on different image datasets show that, in various post-graph learning tasks (i.e., Label propagation and Manifold learning), the graph which is constructed by the proposed method, even after hundreds of insertions and updates, has a similar performance (in some cases it can be better) compared to the graph which is constructed from scratch. (C) 2019 Elsevier Ltd. All rights reserved.
Keyword:
Incremental graph construction
Locality-constrained linear coding
Graph-based label propagation
Graph-based linear manifold learning
Semi-supervised learning
Face recognition
AI总结
对已上传原文的论文进行重点信息的提取，主要内容包括：简要概述、研究摘要、背景介绍、关键亮点、图文解析、展望与总结。
期刊
IF:
7.5
论文数:
3.0W
被引数:
10.2W
机构
引用论文
.png){kind=icon}
Effects of stereoisomers of estradiol on food intake, body weight and hoarding behavior in female rats雌二醇立体异构体对雌性大鼠摄食、体重和囤积行为的影响
PREAI
分享收藏A Review of Technical Impact of Electrical Vehicle Charging Stations on Distribution Grid电动汽车充电站对配电网的技术影响研究综述
PREAI
分享收藏Mechanical characterization of neural electrodes based on PDMS-parylene C-PDMS sandwiched system
PREAI
分享收藏分享收藏
查看更多内容