Computation of two-layer perceptron networks’ sensitivity to input perturbation

The sensitivity of a neural network’s output to its input perturbation is an important measure for evaluating the network’s performance. In this paper we propose a novel method to quantify the sensitivity of a Two-Layer Perceptron Network (TLPN). The sensitivity is defined as the mathematical expectation of absolute output deviations due to input perturbations with respect to all possible inputs. In our method a bottom-up way is followed, in which the sensitivity of a neuron is first considered and then is that of the entire network. The main contribution of the method is that it requests a weak assumption on the input, that is its elements need only to be independent identically distributed, and thus is more practical to real applications. Some experiments have been conducted, and the results demonstrate high accuracy and efficiency of the method.