Please enable JavaScript.
Coggle requires JavaScript to display documents.
Introduction (A Primer on Neural Networks (The role of the neuron (Each…
Introduction
A Primer on Neural Networks
Especially useful on
have lots of historical or sample data, but to which hard and fast rules cannot be easily applied
problems which are tolerant of some error
Constructed of inter connected nodes
Can Calculate on any computable function
Input Layer
Hidden Layer
Output Layer
Came out of a desire to simulate physiological structure of human brain
At their simplest, feed forward nn propagate attribute information through the nn to make a prediction, whose output is either continuous for regression and discrete for classification.
The role of the neuron
Take input information
Process via mathematical function
distributes to hidden payer neurons
Activate each other via weighted sums
Each neuron contains
Activation function
Threshold value
Activation Functions
Neural Network Learning
Original "Preceptron" model developed at Cornell Aeronautical Laboratory in 1958
Output Layer
Association unit combines input with weights and a threshold step function
Retina - distributed inputs to second layer
Who Uses Deep Learning
Microsoft
Google
Natural Language Processing
Facebook
Medical Image Processing
What Problems can Deep Learning Solve
Classify or Predict nonlinear data using modest number of parallel nonlinear steps
What is Deep Learning
Deep Learning Pyramid
Multi-layer nonlinear models
Unsupervised Learning
Training data does not contain known outcomes.
Algorithm self-discovers relationships in data
Neural Networks
Supervised Learning
Training data contains known outcomes
Model is trained relative to those outcomes
General Deep Learning Framework
Input Data
Layer 1
Layer 2
Layer k
Simple Classifier
Object Class
