As part of my coursework at NTU, I will be doing research on the topic of Semi-Supervised Clustering with Graph Neural Networks under Prof Xavier Bresson.
In graph clustering, the goal is to predict for each node in a graph, the cluster that it belongs to. In semi-supervised graph clustering, we are given as input a single node from each cluster, along with the entire graph. Unlike clustering in the unsupervised setting, we know exactly how many clusters there are, and the position of a random node from each of these clusters.
There will be a series of upcoming blogposts detailing my goals, progress and understanding gained on this topic.
In particular, my research would be geared towards examining, comparing and benchmarking in the semi-supervised clustering setting, three recent variants of graph neural networks, including:
X. Bresson and T. Laurent. Residual Gated Graph ConvNets. arXiv preprint arXiv:1711.07553.
P. Veličković, G. Cucurull, A. Casanova, A. Romero, P. Liò and Y. Bengio. Graph Attention Networks. ICLR 2018.
T. Kipf and M. Welling. Semi-Supervised Classification with Graph Convolutional Networks. ICLR, 2017.