Sie sind hier

Estimating K-Core Numbers using Transfer Learning

The k-core number of a node is an essential centrality metric in network analysis [1]. However, performing k-core decomposition to the entire network is unrealistic in many real-world scenarios. This raises the need to estimate k-core numbers using only nodes’ local information [2]. Transfer learning, enables us to learn knowledge from familiar domains, and transfer to other domains [3]. We aim to use transfer learning, in order to learn a model from networks in which k-core numbers of nodes are known (or can be computed), and apply this model to estimate k-core numbers of nodes in other networks where performing k-core decomposition is unrealistic. In this thesis, the student reads and studies related literatures on k-core decomposition [1, 4], k-core number estimation [2] as well as transfer learning [3]. The student implements a transfer learning algorithm which learns a model from (an) existing network(s) and estimates nodes’ k-core numbers in other networks without having the entire network as input. The student then applies the algorithm to real-word networks and evaluates its performance (accuracy and time complexity) with existing approaches.

Prerequisites: Python or Matlab/Octave, knowledge in network theory and machine learning.

[1] F. D. Malliaros, A. N. Papadopoulos, and M. Vazirgiannis, “Core decomposition in graphs: Concepts, algorithms and applications,” in EDBT, 2016, pp. 720–721.
[2] M. P. O’Brien and B. D. Sullivan, “Locally estimating core numbers,” in 2014 IEEE International Conference on Data Mining (ICDM), IEEE, 2014, pp. 460–469.
[3] S. J. Pan and Q. Yang, “A survey on transfer learning,” IEEE Transactions on knowledge and data engineering, vol. 22, no. 10, pp. 1345–1359, 2010.
[4] W. Khaouid, M. Barsky, V. Srinivasan, and A. Thomo, “K-core decomposition of large networks on a single PC,” Proceedings of the VLDB Endowment, vol. 9, no. 1, pp. 13–23, 2015.