|
Kategorie: Diplomové, bakalářské práce |
Tento dokument chci!
Práce popisuje základy principu funkčnosti neuronů a vytvoření umělých neuronových sítí. Je zde důkladně popsána struktura a funkce neuronů a ukázán nejpoužívanější algoritmus pro učení neuronů. Základy fuzzy logiky, včetně jejich výhod a nevýhod, jsou rovněž prezentovány. Detailněji je popsán algoritmus zpětného šíření chyb a adaptivní neuro-fuzzy inferenční systém. Tyto techniky poskytují efektivní způsoby učení neuronových sítí.
Algorithm itself includes three periods: feedforward spreading the
input signal training pattern, backward spreading errors and actualization of
weighted values connections.10
Backpropagation algorithm used approximately 80% all neural network
applications.
What needed for learning the neural network? both the training set
containing elements describing the solved problem and then method that can fix
these samples the form neural network synaptic weight values, including the
already mentioned ability generalize, possible., Zp). Signal spreading biological
system proceeds such way too, where input layer can created e. Stop first the training set.1)
Ti [t1 tm] ‹0, 1›
where number training set patterns
Si excitation vector the input layer consisting neurons
Ti excitation vector the output layer consisting neurons
sj, excitation the jth
neuron the input, respectively the output
layer
The method that allows the adaptation the neural network training set is
called backpropagation.,n) receives input signal (xi) and mediates its transfer all neurons the inner
layer (Z1.
During feedforward signal spreading, each neuron the input layer (Xi, =
1,.. The process determining the synaptic
weights linked again with the concept learning the neural networks.
Formally, for the training set can consider set elements (patterns) that are
arranged pairs defined follows:
T {{S1,T1} {S2,T2}…{Sq,Tq}}
Si [s1 sn] ‹0, (3.g... Each neuron the inner layer calculates its activation (zj) and sends
this signal all the neurons the output layer. This method adaptation the opposite direction the
spread information from higher layers lower layers. Each
training set pattern describes, how neurons are excited the input and output layers.
In principle this way response neural net the input stimulus can be
obtained, given excitation input layer neuron. with visual
cells and the output layer the brain are then identified individual objects of
watching.
Another issue the ability generalization over the learned material, other
words, how the neural network able deduce the basis learned phenomea
that were not part the learning process, but can somehow deduced from the
learned. Each neuron the output layer
calculates its activation (yk), which matches its real output (kth
neuron) after
submission the input sample. The question then will be, how synaptic weights leading correct
response the input signal are defined.
.
