Evolving graph convolutional networks for dynamic graphs

A non-transitory computer readable medium comprising computer executable instructions which when executed by a computer cause the computer to perform the method of:

• for each of a plurality of graph convolutional networks corresponding to a plurality of time steps, each graph convolutional network modelling a graph including nodes and edges, each graph convolutional network comprising a plurality of graph convolution units, for a given one of said time steps, taking as input a graph adjacency matrix, a node feature matrix, and a parameter matrix for a current layer, and outputting a new node feature matrix for a next highest layer;
• with an evolving mechanism, taking as input a parameter matrix for a time step prior to said given one of said time steps, updating said input parameter matrix, and outputting said parameter matrix for said given one of said time steps;
• and
• obtaining, as input to an output layer in data communication with said evolving mechanism and said plurality of graph convolutional networks, output of each of said graph convolution units for a final one of said time steps, and based on said output of each of said graph convolution units for said final one of said time steps, outputting from said output layer a graph solution output.

A method comprising:

• for each of a plurality of graph convolutional networks corresponding to a plurality of time steps, each graph convolutional network modelling a graph including nodes and edges, each graph convolutional network comprising a plurality of graph convolution units, for a given one of said time steps, taking as input a graph adjacency matrix, a node feature matrix, and a parameter matrix for a current layer, and outputting a new node feature matrix for a next highest layer;
• with an evolving mechanism, taking as input a parameter matrix for a time step prior to said given one of said time steps, updating said input parameter matrix, and outputting said parameter matrix for said given one of said time steps;
• and
• obtaining, as input to an output layer in data communication with said evolving mechanism and said plurality of graph convolutional networks, output of each of said graph convolution units for a final one of said time steps, and based on said output of each of said graph convolution units for said final one of said time steps, outputting from said output layer a graph solution output.