Luca Cosmo et al. propose in this work to extend the classic convolution operator to the graph domain by using #graphkernels, i.e. #kernelfunctions that compute inner product on graphs.
This defines a complete structural model without having to compute the input graph's embedding
Time with fancy, science-kit for machine learning (#sklearn). Here's a neat exercise comparing ML model accuracy to classify & predict loan statuses based on a history file; imp visualizing variables (#K, #TreeDepth, & todo #KernelFunctions & #Regularizat…lnkd.in/er_92bi