MbShea@PWP@English@UMCP

ABSTRACT: In this tutorial, the motivation behind deep neural networks and applications of deep models are covered. The contemporary functions of networks in self-driving cars, image classification and recognition, fake news generation, and artistic style transfer are expanded upon. After this brief exposition, the author starts off by reviewing some foundational work in the theory of neural networks, such as training these networks through backpropogation, as well as the first applications of networks for digit recognition for postal zip codes. The building blocks of neural network architectures are elaborated, with detail given into convolutional layers, fully connected layers, and activation functions. Recent advances in the field are also given consideration, such as the ReLu activation function and dropout layers in networks. After the general intuition of the layers and supporting building blocks of the networks are described, the author then constructs a neural network using the PyTorch library to seam the concepts together in an example to tackle the MNIST handwritten digit recognition problem.

WC=163

READER’S PROFILE: I can imagine a reader being overwhelmed just by the term neural network. They may assume there is rigorous mathematics involved, which may not be accessible to them.

READER’S RESPONSE: Wow. Neural networks are pretty hot right now if they are being used in all these fields of industry and research. I wonder what kind of problem that I can use a neural network for. But the coding looks daunting… How do I even know what neural network I should use for the problem I want to tackle? Maybe I should do more research into the topic since I’m interested… Where can I learn more about the theory behind these networks? What are the state-of-the-art advances in the field?

R, looks like a good plan. You can ADD the logos of numbers to the applications area:

project economics of AVs
image recognition, including in security.

And, for fun and learning, I would include the case of Janelle Shane who trains networks for fun. Check out her website:

http://aiweirdness.com/

I love her project on knitting patterns, by AI. Also, I am trying to assemble a team of students who will scrape to build a N.N. database to train on common plant names.