4. AGENDA
What is a Neural Network
History of Neural Networks
Types of learning for Neural Networks
Where are Neural Networks applicable
Neural Networks vs Conventional Computers
5. AGENDA
What is a Neural Network
History of Neural Networks
Types of learning for Neural Networks
Where are Neural Networks applicable
Neural Networks vs Conventional Computers
6. AGENDA
What is a Neural Network
History
The term neural network was traditionally Types
used to refer to a network or circuit Where
of biological neurons. Neural
An artificial neural network, is a mathematical
model or computational model.
What has attracted the most interest in neural
networks is the possibility of learning. Given a
specific task to solve.
7. AGENDA
History of Neural Networks
What
1940's - The first artificial neuron was
The term neural network was traditionally Types
produced by the to a network or Warren
used to refer neurophysiologist circuit Where
McCulloch and the logician Walter Pits.
of biological neurons. Neural
1950's - The perceptron by Frank
An artificial neural network, is a mathematical
Rosenblatt.
model or computational model.
1980's - D.O.D, Boltzmann machines,in neural
What has attracted the most interest
Hopfield nets, competitive learning models, a
networks is the possibility of learning. Given
multilayer networks.
specific task to solve.
8. AGENDA
Types of learning for Neural Networks
Supervised learning What
1940's - wants firstinfer thewasneuronimplied
The user The to artificial
The term neural network mapping was
traditionally History
produced by the costa networkrelated to the
by the to refer to function is or Warren
used data, the neurophysiologist circuit Where
McCulloch and the logician Walterand the data
mismatch between our mapping Pits.
of biological neurons. Neural
and it implicitly contains prior knowledge
1950's the problem network, is a mathematical
about - The perceptron by Frank
An artificial neural domain.
Rosenblatt.
model or computational model.
Unsupervised learning
1980's - D.O.D, Boltzmannthe cost function to
Some data is given and machines,
What has attracted the most interest in neural
networks is the The cost function is dependent
be minimized, possibility of learning. Given
Hopfield nets, competitive learning models, a
multilayer networks. the implicit properties of
on the task and on
specific task to solve.
our model, its parameters and the observed
variables.
9. AGENDA
Types of learning for Neural Networks
Reinforcement learning What
1940's - The first not given, neuron was
The data are usually artificial but generated
The term neural network was traditionally History
produced aby n t ' sto i n t e networks or iWarren
useda n refer neurophysiologist tcircuit
by to g e the a raction w h the Where
McCulloch andAt each point in time the agent
environment. the logician Walter Pits.
of biological neurons. Neural
performs an action and the environment
1950'sr- The perceptron by Frank a n d a n
An artificial s a n network,aisi o n
g e n e a t e neural o b s e r v t a mathematical
instantaneous cost.
Rosenblatt.
model or computational model.
Learning algorithms
1980's - D.O.D, Boltzmann machines,in neural
What has attracted the most interest
Hopfield nets, competitive learning models, a
Training a neural network model essentially
networks is the possibility of learning. Given
multilayer networks.
specific task to solve. model from the set of
means selecting one
allowed model that minimizes the cost
criterion.
10. AGENDA
Where are Neural Networks applicable
Reinforcement learning What
1940's - The first not given, neuron was
The data are usually artificial but generated
The term neural network was traditionally History
produced aby n t ' sto i n t e networks or iWarren
useda n refer neurophysiologist tcircuit
by to g e the a raction w h the Types
McCulloch andAt each point in time the agent
environment. the logician Walter Pits.
of biological neurons.
•Investment analysis and the environment Neural
performs an action •Robotics
•e n e r- The perceptrone by Frank a n d a n
Credit Evaluationo b s r v a t i o n
1950's a t e s a n network, is a mathematical
g •Medicine
An artificial neural
•Signature analysis •Weather
instantaneous cost.
Rosenblatt.
model or computational model.
•Marketing •Intelligent Searching
•Monitoring
Learning algorithms •Games
1980's - D.O.D, Boltzmann machines,in neural
What has attracted the most interest
•Staff scheduling network modelmodels,
Hopfield nets, competitive learning essentially
Training a neural
networks is the possibility of learning. Given a
multilayer networks.
specific task to solve. model from the set of
means selecting one
allowed model that minimizes the cost
criterion.
11. AGENDA
Neural Networks vs Conventional Computers
Reinforcement learning What
1940's - The first not given, neuron was
The data are usually artificial but generated
The term neural network was traditionally History
produced aby n t ' sto i n t e networks or iWarren
useda n refer neurophysiologist tcircuit
by to g e the a raction w h the Types
McCulloch andAt each point in time the agent
environment. the logician Walter Pits.
of biological neurons. Where
performs an action and the environment
What do you think?
1950'sr- The perceptron by Frank a n d a n
An artificial s a n network,aisi o n
g e n e a t e neural o b s e r v t a mathematical
instantaneous cost.
Rosenblatt.
model or computational model.
Learning algorithms
1980's - D.O.D, Boltzmann machines,in neural
What has attracted the most interest
Hopfield nets, competitive learning models, a
Training a neural network model essentially
networks is the possibility of learning. Given
multilayer networks.
specific task to solve. model from the set of
means selecting one
allowed model that minimizes the cost
criterion.