Manifold-Learning-and-Graph-Kernels

Third assignment for A.I. course, Prof. Torsello, Ca' Foscari University of Venice, A.Y. 2018/2019

Assignment

Read this article presenting a way to improve the discriminative power of graph kernels.

Choose one graph kernel among

• Shortest-path Kernel
• Graphlet Kernel
• Random Walk Kernel
• Weisfeiler-Lehman Kernel

Choose one manifold learning technique among

• Isomap
• Diffusion Maps
• Laplacian Eigenmaps
• Local Linear Embedding

Compare the performance of an SVM trained on the given kernel, with or without the manifold learning step, on the following datasets:

• PPI
• Shock

Note: the datasets are contained in Matlab files. The variable G contains a vector of cells, one per graph. The entry am of each cell is the adjacency matrix of the graph. The variable labels contains the class-labels of each graph.