Title:
Community Detection in Bipartite Networks with Stochastic Blockmodels
Community Detection in Bipartite Networks with Stochastic Blockmodels
Speaker:
顏子祺 (Tzu-Chi Yen), Graduate student, Department of Computer Science, University of Colorado Boulder
顏子祺 (Tzu-Chi Yen), Graduate student, Department of Computer Science, University of Colorado Boulder
Time:
11/07 (Sat.) 5 pm PST, 6 pm MST, 7 pm CST, 8 pm EST
11/08 (Sun.) 9 am Taiwan
11/07 (Sat.) 5 pm PST, 6 pm MST, 7 pm CST, 8 pm EST
11/08 (Sun.) 9 am Taiwan
Keywords:
computer science, theoretical computer science, algorithms, statistical physics, Bayesian inference, complex networks, stochastic blockmodel
computer science, theoretical computer science, algorithms, statistical physics, Bayesian inference, complex networks, stochastic blockmodel
Abstract:
The stochastic blockmodel (SBM) is a highly flexible generative model for community detection in complex networks. Previous work has introduced a Bayesian nonparametric formulation to find communities that best compress the model and data. However, although this approach applies in principle to the bipartite network, selecting the number of communities becomes computationally prohibitive because the model requires an additional parameter for each type of vertex. In this talk, I will introduce a prior distribution that respects this bipartite structure and a corresponding algorithm to efficiently and parsimoniously choose the number of communities. This bipartite SBM improves community detection results over general SBMs when data are noisy, improves the model resolution limit, and expands our understanding of the complicated optimization landscape associated with community detection tasks. On a related note, I will introduce some endeavors that promote network science in Taiwan and worldwide.
The stochastic blockmodel (SBM) is a highly flexible generative model for community detection in complex networks. Previous work has introduced a Bayesian nonparametric formulation to find communities that best compress the model and data. However, although this approach applies in principle to the bipartite network, selecting the number of communities becomes computationally prohibitive because the model requires an additional parameter for each type of vertex. In this talk, I will introduce a prior distribution that respects this bipartite structure and a corresponding algorithm to efficiently and parsimoniously choose the number of communities. This bipartite SBM improves community detection results over general SBMs when data are noisy, improves the model resolution limit, and expands our understanding of the complicated optimization landscape associated with community detection tasks. On a related note, I will introduce some endeavors that promote network science in Taiwan and worldwide.
No comments:
Post a Comment