Can we use CNN to effectively learn from a table whose various rows are permuted in the testing dataset, output for each remains same?

Question

I have a set of classes, 37 to be precise. Each class is represented by a feature vector of size [1,10]. A single input sample has the dimensions of [6,10]