Artificial Intelligence: Presentations
We present Deep Graph Kernels, a unified framework to learn latent representations of sub-structures for graphs. Our framework is inspired by latest advancements in language modeling and deep learning. We demonstrate instances of our framework on three popular graph kernels, namely Graphlet kernels, Weisfeiler-Lehman kernels, and Shortest-Path graph kernels. Our experiments on several benchmark datasets show that Deep Graph Kernels achieve significant improvements in classification accuracy over state-of-the-art graph kernels.
Every two-player zero-sum game has an optimal strategy that can be found by solving an LP. But this approach is not polynomial time when the number of pure strategies for each player is exponential in the input, e.g. each player plays spanning trees of a graph. We present fast algorithms to compute Nash-equilibria for these games for bilinear payoff functions using ideas from convex and combinatorial optimization and machine learning.