artificial neural networks (ann’s)

14
Artificial Neural Networks (ANN’s) Jacob Drilling & Justin Brown

Upload: iolana

Post on 23-Feb-2016

64 views

Category:

Documents


1 download

DESCRIPTION

Artificial Neural Networks (ANN’s). Jacob Drilling & Justin Brown. What is an Artificial Neural Network?. A computational model inspired by animals’ central nervous systems. Composed of connected processing nodes (neurons). - PowerPoint PPT Presentation

TRANSCRIPT

Page 1: Artificial Neural Networks  (ANN’s)

Artificial Neural Networks (ANN’s)

Jacob Drilling&

Justin Brown

Page 2: Artificial Neural Networks  (ANN’s)

What is an Artificial Neural Network?• A computational model inspired by animals’

central nervous systems.• Composed of connected processing nodes

(neurons).• They are capable of machine learning and are

exceptional in pattern recognition.• A Network is application specific.

Page 3: Artificial Neural Networks  (ANN’s)

History•Warren McCulloch and Walter Pitts

• Threshold Logic•Frank Rosenblatt

• Perceptron•Marvin Minsky and Seymour Papert

• The Society of Mind Theory•Paul Werbos

• Backpropagation•David E. Rumelhart and James McClelland

Page 4: Artificial Neural Networks  (ANN’s)

Biological Neural Networks• A human neuron has three parts: the cell

body, the axon and dendrites.• The process of sending a signal...

Page 5: Artificial Neural Networks  (ANN’s)

Artificial Networks● The Artificial model is comprised of many

processing nodes (neurons).● Nodes are highly connected with weighted

paths.● It has 3 layers:

○ Input○ Hidden○ Output

Page 6: Artificial Neural Networks  (ANN’s)

Artificial Networks● Each node does its own

processing.● Nodes output according to

their activation function.● Initial weights are random.● Back Propagation Algorithm

“teaches” by changing weights.

 

 

Page 7: Artificial Neural Networks  (ANN’s)

Types

• Functiona. Feed Forwardb. Feed Back

• Structurea. Bottleneckb. Deep learning

Page 8: Artificial Neural Networks  (ANN’s)

Current Uses• Recognition

• Image• Speech• Pattern• Character

• Compression• Image• Audio/Video

• ALVINN - Driverless car

Page 9: Artificial Neural Networks  (ANN’s)

Feed Forward Algorithm• Input -> Output• Each neuron must

sum the weighted products from the previous layer.

• Output using activation function.

• 

Page 10: Artificial Neural Networks  (ANN’s)

Back Propagation•Output -> Input•Training Algorithm•Calculates Error in the output layer

•Propagates Error backwards to change weights

 

Page 11: Artificial Neural Networks  (ANN’s)

Criticism/Negative Aspects• Large amounts of computing power and

storage are needed• Cost efficiency• Human abilities

• Instinct• Logic

Page 12: Artificial Neural Networks  (ANN’s)

Character Recognition

1. Image Processing

1 1 0 0 0 0 0 0 1 11 0 1 1 1 1 1 0 0 10 1 1 1 1 1 1 1 0 10 1 1 1 1 1 1 1 1 00 1 1 1 1 1 1 1 1 00 1 1 1 1 1 1 1 1 01 0 1 1 1 1 1 1 1 01 0 1 1 1 1 1 1 0 11 1 0 0 1 1 1 0 1 11 1 1 0 0 0 0 1 1 1

10

Page 13: Artificial Neural Networks  (ANN’s)

Character Recognition

1

1

1

100..

N = 15

..

N = 10

..1

.

.

0

0

2. Input data in the ANN

Page 14: Artificial Neural Networks  (ANN’s)

Character Recognition

1

1

1

100..

N = 15

..

N = 10

..1

.

.

0

0

2. Input data in the ANN