Events / Topological Limitations of Single Layer Perceptron Networks

Topological Limitations of Single Layer Perceptron Networks

October 13, 2017
12:00 pm - 1:00 pm

University of Pennsylvania DRL A8

Speaker: Darrick Lee

Abstract:

A perceptron network is one of the simplest artificial neural networks, in which the activation function of each neuron is a step function. We impose the additional restriction that each neuron can only have a fixed number of inputs. Given an N x N binary image (all pixels either black or white), can such a network detect whether or not the black pixels are connected? The answer is no (kind of).