Extensive Deep Neural Networks (EDNN) were first introduced by my coworkers and me in “Extensive Deep Neural Networks” (https://arxiv.org/abs/1708.06686). The EDNN technique of training and…

# Category: EDNN

Extensive Deep Neural Networks

An obvious extensive property is the number of visible objects within an image. We will tackle an over-simplified model of the common carnival game where…

In this post we’ll step up our game a bit and move on to a slightly more challenging problem: the Ising model of magnetization. This…

So far, we’ve built an EDNN using arbitrary choices for focus and context sizes. Unfortunately this choice is largely based on trial and error, but…

Arguably, the most important feature of extensive deep neural networks is a trained model’s transferability to input sizes larger than those on which it was…

EDNNs can also be implemented in PyTorch. See the GitHub repo (https://github.com/millskyle/extensive_deep_neural_networks/blob/master/ednn_blog_2_ising-pytorch.ipynb) for the Ising Model tutorial (Part 2) reimplemented in PyTorch. It even beats…