Download Artificial Neural Networks - Computer Science, Stony Brook University

Survey
yes no Was this document useful for you?
   Thank you for your participation!

* Your assessment is very important for improving the workof artificial intelligence, which forms the content of this project

Document related concepts

History of artificial intelligence wikipedia , lookup

Pattern recognition wikipedia , lookup

Catastrophic interference wikipedia , lookup

Hierarchical temporal memory wikipedia , lookup

Convolutional neural network wikipedia , lookup

Transcript
Artificial Neural Networks
Group #4
John Dilag 108613685
Eric Loo 108818998
Jessica Zeng 108591189
Anita Wasilewska
CSE 352: Artificial Intelligence
References
[1] http://www.psych.utoronto.ca/users/reingold/courses/ai/cache/neural2.html
[2] http://cdn.grid.fotosearch.com/LIF/LIF114/sa302026.jpg
[3] https://www.willamette.edu/~gorr/classes/cs449/ann-overview.html
[4] https://www.youtube.com/watch?v=DG5-UyRBQD4
[5] https://en.wikibooks.org/wiki/Artificial_Neural_Networks/Neural_Network_Basics
[6] http://www.cs.utsa.edu/~bylander/cs6243/neural-networks.pdf
[7] http://www.cob.calpoly.edu/~eli/pdf/neural.pdf
[8] http://www.slideshare.net/Ahmed_hashmi/neural-network-its-applications
[9]http://cs.stanford.edu/people/eroberts/courses/soco/projects/neural-networks/History/history1.html
[10] http://psych.utoronto.ca/users/reingold/courses/ai/cache/neural4.html
[11] http://www.alyuda.com/products/forecaster/neural-network-applications.htm
[12] http://citeseerx.ist.psu.edu/viewdoc/download?doi=10.1.1.165.8217&rep=rep1&type=pdf
2
Overview
1. What are Artificial Neural
Networks?
2. History
3. How do they work?
4. How do they learn?
5. Applications of ANNs
6. Conclusion
3
Artificial Neural Networks
Why should we
care?
4
5
What are Artificial Neural Networks (ANN)?
•  Part of machine learning and cognitive
science
•  Modeled after the information processing
and physical structure of neuron
connections in our brain to solve problems
that cannot be solved by traditional
computing
•  Capable of generalization - the ability to
recognize similarities among different input
patterns (pattern recognition)
•  Can be trained to solve certain problems
using a teaching method and sample data
Sources: https://en.wikibooks.org/wiki/Artificial_Neural_Networks/Neural_Network_Basics [5], https://shillyard.files.wordpress.com/2013/02/brain-rewired1.jpg
6
History of Neural Networks
• 
• 
• 
• 
• 
• 
• 
• 
• 
• 
1943 - Warren McCulloch and Walter Pitts; Paper on how neurons work, Modeled a simple Neural
Network
1957/8 - Frank Rosenblatt, Charles Wightman and others developed the first successful neurocomputer, Mark I perceptron
1959 - ADALINE, new learning laws and binary pattern recognition and MADALINE an adaptive
filter that eliminates echoes from phone lines
1975 - First multilayered network
1982 - John Hopfield renewed interest in the topic of NN
1986 - Publication of PDP (Parallel Distributed Processing) books sparked boom in NN
1987 - The IEEE annual international ANN conference was started for ANN researchers.
1988 - The International Neural Network Society (INNS) journal was founded
1989 - Neural Computation journal
1990 - IEEE Transactions on NN journal
Sources: http://cs.stanford.edu/people/eroberts/courses/soco/projects/neural-networks/History/history1.html, http://www.springer.com/cda/content/document/cda_downloaddocument/
9789401798150-c2.pdf?SGWID=0-0-45-1495021-p177264210
7
How do ANN work?
•  Neurons provides us with our abilities to remember, think, and apply previous
experiences to our every action
•  Humans have many variations of neurons, but they all have the same four basic
components:
o  Dendrites
o  Soma
o  Axon
o  Synapses
Sources: http://www.psych.utoronto.ca/users/reingold/courses/ai/cache/neural2.html
8
How do ANN work: Neuron
Sources: http://cdn.grid.fotosearch.com/LIF/LIF114/sa302026.jpg
9
How do ANN work: Artificial Node
Each node takes in various inputs, and each input is multiplied by its associated
weight, wi. The products are them summed, fed through a transfer function to set
its value, and then outputted to the next set of nodes.
Sources: http://www.psych.utoronto.ca/users/reingold/courses/ai/cache/neural2.html, https://www.willamette.edu/~gorr/classes/cs449/ann-overview.html
10
How do ANN work: Layers
11
How do they learn?
Learning for ANN is finding values of weight connections that minimizes error by using backpropagation:
1.  Initialize with random connection weights
2.  Feed the training sample - set of given input and desired output
3.  Let the network calculate the output with the given inputs forward
4.  Calculate the error
5.  Output nodes communicate to the hidden nodes the error and each pair adjusts the connection
weights between. The nodes continue to push back until all nodes have been assigned an error
6.  Network will try the original inputs again, and repeat this process until the output is desired
Sources: https://www.youtube.com/watch?v=DG5-UyRBQD4, http://www3.cs.stonybrook.edu/~cse352/L12NN.pdf
12
Applications of ANNs
ANNs are used in many applications in the recent years. Some examples include but are not limited to
• 
Facial Recognition
• 
Written number and letter Recognition
• 
Stock market predictions
• 
Staff Scheduling
• 
Fraud Detection
• 
Pattern Recognition
• 
College Application Screenings
• 
Fingerprint Recognition
Sources:http://www.alyuda.com/products/forecaster/neural-network-applications.htm
13
Applications of ANNs: Written Number Recognition
Sources: http://citeseerx.ist.psu.edu/viewdoc/download?doi=10.1.1.165.8217&rep=rep1&type=pdf
14
Applications of ANNs: Written Number Recognition
Sources: http://citeseerx.ist.psu.edu/viewdoc/download?doi=10.1.1.165.8217&rep=rep1&type=pdf
15
Applications of ANNs: Written Number Recognition
https://www.youtube.com/watch?
v=QmNydKGBlxQ
The size of the training sample of 5000 images.
The total number of weights in a network 158 000.
Digitize
Checks
Fillable Forms
Papers
Documents
This fully-meshed network consists of three layers.
The first layer contains 196 elements.
The second layer contains 25 elements.
The output layer includes ten elements
Sources: https://www.youtube.com/watch?v=QmNydKGBlxQ
16
Future In ANN
Recent experimental data has provided further evidence that biological neurons
are structurally more complex than the simplistic explanation above. They are
significantly more complex than the existing artificial neurons that are built into
today's artificial neural networks. As biology provides a better understanding of
neurons, and as technology advances, network designers can continue to improve
their systems by building upon man's understanding of the biological brain.
Sources: http://www.psych.utoronto.ca/users/reingold/courses/ai/cache/neural2.html
17
Thank you!
18