The theory of resonance adaptive (or Adaptive Resonance Theory) by Stephen Grossberg and Jill Carpenter with a view to how information processing by the brain, etc. were raised. This theory of a few neural network model to describe track in which the methods of regulatory and non-regulatory, to learn used. This theory, in issues such as diagnosis and prediction of patterns, etc. to be handled.

The System, ART Elementary, a learning model derelict is. This system, usually a field comparison, and a field detection (made up of a number of neurons), a parameter of the caregiver (vigilance parameter), and a unit Reset (Reset Module) is made. Parameter careful usually is significant effect on the system is: if the value of this parameter in the considered memory that with high detail (the floor of small, but high) at the discretion of puts and vice versa, if this small amount of. memory of normal (the floor of the regular size and a low number of) gotten gives.[ field of comparison, a carrier input (an array of one-dimensional quantities) received, and it is the best in the field recognize the transferred track. This is the best peer., the neuronal lone that sets the weight of his (Weight Vector), with carrier input, the greatest match will have. Every neuron in field detection of a negative signal (proportional to the quality of the match carrier, input with the neuron receiving it) to other neurons, the field post is ... the result of the production output in the block can be. With this, field detection methods, Contraindications, Side does offer. This means that the neurons inside it, as a class (categories) acts which the carrier input based on the categories are. After input, was classified as the unit resets, the power of this conformity, and categories with the parameter value watch out for cons. If the answer of the tests was positive., the process of teaching starts. On the other hand, if answer comparing the negative is. neuronal that the signal is sent. the received carrier, the new entry becomes disabled; the process of education, too, only after complete the search process begins. In the process of search, the neurons, the field detection one by one by the unit reset disabled, are, until, finally, the result of a measurement, positive be. But if the power none of the compliance has been, to the threshold parameter value, be careful not (the response measure is always negative in.) a neuron that had to be handled not to be launched, and so adjusted to the carrier input match.

the two main methods for training neural networks based on ART there:one method quiet and the other quick method. In the method of learning, quiet, etc. from differential equations to calculate the values of the Continuous of the degree of weight sharing neurons detect relative to the carrier input, etc. used; therefore, this calculation to the period of time during which the carrier input is provided depends. In the method of learning fast, from algebraic equations to calculate the grade adjustments weight, etc. can be used; and the values for binary are. Although the learning method fast, effective, and efficient. the learning method, relaxed, etc. in terms of biodiversity is more likely and can be used in networks with continuity when the (continuous-time networks) to be employed (for example, when the carrier input, periodically change).