Title: How we discovered Deep Neural Networks at Middlesex University in 2007.
Abstract
We represent neural networks by directed graphs and consider the
problem of maximal connectivity with constraints. This problem is
motivated by some conflicting objectives in the design of biological
neural networks. One of the solutions of this optimisation problem
of neural network with constrains was presented at IJCNN 2007 and
resembles what is now called a concolutional deep-learning neural
networks. Inequalities and equations derived are tested on data and
numerical estimates for parameters of a human brain. Results support
an intuition that human brain is maximally connected subject to
constraints on in- and out-degrees.