edgeconv

one set of particlenet like update steps, thus use m.edge* like subedge. also similar to gq (here the main difference is the dense vs convolutional structure

Arguments:

• inp: input variable
• gs: node number
• k: the k in topK
• param: number of parameters
• m: constant variable, created for example by getm()

Returns

• outp: the updated feature vector