Neural Networks
from the Ground up
Frank La Vigne
FranksWorld.com / DataDriven.tv / @tableteer /
FranksWorld.tv
FrankLa@microsoft.com
NeuralNetworks
Whatdotheyknow?
Dotheyknowthings??
Let’sFindOut!
DataSciencePodcast
Data Driven Podcast
 Launched June 2017
 178 Episodes
 Over 80,000 downloads
 iTunes, Google Play, Spotify, Tune In, Facebook, etc
Writing
 Regular column in MSDN Magazine
 aka.ms/FrankMSDN
 Two new books on Data Science
 Yes, two
 Yes, I’m quite possibly insane
UsuallyYouSeeThis
 What is going on in this diagram?
 How does this structure work?
 Why does this structure work?
Inspired by Biology
CloserLookataNeuron
Santiago Ramón yCajal
 Santiago Ramón y Cajal (1852-1934) ventured
into science as both an artist and a pathologist
and became the first person to see a neuron.
 Working by gaslight, he made thin slices of brain
tissue and subjected them to the same silver-
nitrate chemistry he used to capture images on
photographic plates.
Starting toLookFamiliar
Artificial Neural Networks
 Modeled after biological neural networks
 Activation functions
 Pros
 Amazing results
 Cons
 Lack of explainability
 Computationally expensive
 Complexity
ArtificialNeuron
Sigmoid
Rememberthis
Graphic
CodeTime!
http://bit.ly/frankadfcode
https://notebooks.azure.com/FranksWorld/projects/Neural
NetworksBasics
CONVOLUTIONAL NEURAL NETWORK
dog
bicycle
apple
tennis
ball
CONVOLUTIONAL NEURAL NETWORK
dog
bicycle
apple
tennis
ball
TRANSFER LEARNING
TRANSFER LEARNING
hotdog
not
hotdog
F1
F2
F3
F4
F5
F6
F7
F8
Questions?
Recommended Training
Contactme
 frankla@Microsoft.com
 @tableteer
 FranksWorld.com
 DataDriven.tv
 aka.ms/FrankMSDN

Neural Networks from the Ground Up