Perceptron (P)

Recognition of characters printed

The oldest network

You may not recognize complex characters

1958

Rosemblatt

Adaline/Madaline

Filtering of signals. Equalizer Adaptive

Quick, easy implement with analog circuits

It is only possible classifying spaces linearly separated

1960

Widrow

Avalanche

Recognition of talks continued. Robot control

No network easy done these tasks

It is not easy to alter the speed or interpolate the movement

1967

Grossberg

Cerebellation

Control of the movement of arms of a robot

Similar to Avalanche

It requires complicated control inputs

1969

Marr,

Albus,

Brain-State-Box

Extraction of knowledge database

Possibly best realization that SAS Hopfield networks

Potential realization applications not completely studied

1977

Anderson

Neocognitron

Recognition of characters manuscripts

Easy conceptualize. Insensitive to the translation, rotation and scale

It requires many elements of process, levels and connections

1978

Fukushima

Self-Organizing-Map (SOM)

Recognition of patterns, coding data optimization

Make maps features common of cough learned data

It requires a lot training

1980

Kohonen

Hopfield

Reconstruction patterns and optimization

Can be implemented in VLSI.

Capacity and stability

1982

Hopfield

Machines Boltzmann and Cauchy

Recognition patterns (images, sound and radar). Optimization

Simple networks.

Capacity of representation optimal patterns

The machine of Bolzmann need a very long time learning

1985

Hinton

Associative memory bi-directional

Memory heteroasociativa access by content.

Learning and simple architecture

Low capacity storage. The data should be encoded

1985

Kosko

Adaptive resonance theory (ART)

Recognitionpatterns (radar, (sonar, etc.)

Sophisticated, little used

Sensitive to the translation, distortion and scale

1986

Carperrter, Grossberg

Counterpropagation

Understanding images

Combination of Perceptron and TPM

Many neurons and connections

1986

Hectnielsen

Back-propagation

(BP)

Voice synthesis from text. Control robots. Forecast Recognition patterns

The most popular network. Numerous applications successfully. Ease of learning. Powerful.

It needs a lot of time for the learning and many examples

1986

Werbos, Parker,

Rumelhart