75. Topology classification using a Multiple Kernel Learning approach with graphs and non-graphable data
Contributed abstract in session FB-2: Machine learning, stream Machine learning.
Friday, 10:30 - 12:00Room: M228
Authors (first author is the speaker)
| 1. | Maciej Majchrzak
|
| Poznan University of Technology | |
| 2. | Piotr Lukasiak
|
| Institute of Computing Science, Poznan University of Technology |
Abstract
In this study, the possibility of classifying topologies described by both graphable and non-graphable data using kernels was investigated. The focus was on 3D shape classification - especially CAD geometry models - using graph kernels coupled with Support Vector Classification.
Apart from the graphs describing the topology, each shape was also characterized by a number of parameters that cannot be depicted using graphs. A simple method was developed to incorporate these parameters into the classification process without any modification to the tools used in the aforementioned graph classification. In addition, the possibility of using Multiple Kernel Learning for this purpose was tested. The results show that supporting graph classification with additional data, even carried out in the simplest way, can noticeably improve classification quality. It also proved crucial to understand the meaning of data to avoid the overestimation of classification accuracy.
Keywords
- Machine learning
- Graph theory and networks
Status: accepted
Back to the list of papers