Abstract | ||
---|---|---|
Graph costructing plays an essential role in graph based learning algorithms. Recently, a new kind of graph called L1 graph which was motivated by that each datum can be represented as a sparse combination of the remaining data was proposed and showed its advantages over the conventional graphs. In this paper, the L1 graph was extended to kernel space. By solving a kernel sparse representation problem, the adjacency and the weights of the graph are simutaneously obtained. Kernel L1 graph preserved the advantages of L1 graph and can be more robust to noise and more data adaptive. Experiments on graph based learning tasks verified the supeririority of kernel L1 graph over the conventional K-NN graph, epsilon-ball graph, and the state-of-the-art L1 graph. |
Year | DOI | Venue |
---|---|---|
2012 | 10.1007/978-3-642-33506-8_55 | PATTERN RECOGNITION |
Keywords | Field | DocType |
Kernel Sparse representation,Kernel L1 Graph,Spectral Embedding,Spectral Clustering | Strength of a graph,Discrete mathematics,Line graph,Graph property,Theoretical computer science,Null graph,Butterfly graph,Voltage graph,Mathematics,Graph (abstract data type),Complement graph | Conference |
Volume | ISSN | Citations |
321 | 1865-0929 | 0 |
PageRank | References | Authors |
0.34 | 13 | 4 |
Name | Order | Citations | PageRank |
---|---|---|---|
Liang Xiao | 1 | 39 | 6.64 |
Bin Dai | 2 | 51 | 5.09 |
Yuqiang Fang | 3 | 128 | 8.93 |
Tao Wu | 4 | 58 | 11.53 |