Neural Networks
1
Neural networks
A beautiful biologically-inspired programming which
enables a computer to learn from observational data
* Some content of the slides are taken from “Neural networks and deep learning”
by Michael Nielsen, Determination Press, 2015
2
Neural networks
Neural networks and deep learning currently
provide the best solutions to many problems in
the areas of:
Image recognition
Speech recognition
Natural language processing
3
Conventional programming
versus neural network
Conventional approach to programming
we tell the computer what to do, breaking big
problems up into many small, precisely defined tasks
that the computer can easily perform.
Neural network
we don't tell the computer how to solve our problem.
Instead, it learns from observational data, figuring out
its own solution to the problem at hand.
4
Husky or Wolf ?
5
Recognize hand written digits
Humans have a primary visual cortex, V1,
containing 140 millions neurons, with tens of
billions of connections between them.
Human vision involves not just V1, but an entire
series of visual cortices.
How to write a computer program to recognize
the digits above?
6
Neural network approach
Take a large number
of handwritten digits,
known as training
examples
7
Neural network approach
The neural network
use the examples to
automatically infer
rules for recognizing
handwritten digits.
8
Neural network approach
By increasing the
number of training
examples, the
network can learn
more about
handwriting, and so
improve its
accuracy.
9
Recognize hand written digits
A
computer program implements a neural network
that learns to recognize handwritten digits.
 The
program can recognize digits with an accuracy
over 96 percent.
 The
best commercial neural networks are used by
banks to process checks and by postal offices to
recognize addresses.
10