Text this: A low functional redundancy-based network slimming method for accelerating deep neural networks