| Title |
Comparison Between Brain Sub-Networks Decomposed by Auto Encoder and Graph Auto Encoder with Non-Negative Weight Constraint and Sparse Encoding |
| Authors |
이필섭(Pilsub Lee) ; 최명원(Myungwon Choi) ; 김대겸(Daegyeom Kim) ; 이수지(Suji Lee) ; 윤현철(HyunChul Youn) ; 정현강(Hyun-Ghang Jeong) ; 한철(Cheol E. Han) |
| DOI |
https://doi.org/10.5573/ieie.2019.56.12.99 |
| Keywords |
brain networks ; graph auto-encoder ; sub-network ; decomposition quality ; spatial locality ; |
| Abstract |
Brain networks consist of nodes, which is anatomically defined brain regions, and edges, which connects a pair brain regions, that is strongly associated with cognitive functions. Brain regions dedicated to a certain specific cognitive function are spatially clustered and efficiently connected each other. The recently developed deep learning model for graph data, the graph auto encoder, can successfully decompose the brain networks into such clustered brain regions and their networks, i.e. sub-networks. However, as far as the authors’ knowledge, there has been no clear way to evaluate the decomposition quality. In this study, we propose the sparsity and spatial locality as measures of the decomposition quality, and compared performance of the traditional machine learning technique, auto-encoder and the newly developed deep learning algorithm, graph auto-encoder with respect to these measures. |