Artificial neural network architectures
- 2. Perceptron Network • Weights between input & output units are adjusted. • Weights between sensory associator units are fixed. • Goal of Perceptron net is to classify the input pattern as a member on not a member to a particular class. X1 Xi 1 Xn Y X0 X1 Xi Xn y b W1 W2 Wn
- 3. Adaline Network • Receives input from several units and one unit called bias. • Inputs are +1 or -1, weights have sign + or - • Net input calculated is applied to quantizer function to restore output to +1 or -1 • Compares actual
- 4. Madaline Network • Contains “n” units of input layer,”m” units of adaline layers, “1” unit of Madaline Layer. • Each neuron in the Adaline and madaline layer have a bias of excitation 1. • Adaline layer is present between input and output Madaline Layer. • Used in Communication Systems , equilizers and noise cancellation devices.
- 5. Back Propagation Network • A multilayer Feed forward network consisting of Input, hidden and output layers. • Hidden and output layers have biases whose activation is 1. • Signals are reversed in learning phase. • Inputs sent to BPN and outputs obtained could be
- 6. Auto Associative Memory Network • Training input and target output vectors are same. • Input layers consist of n input units & output layer consist of n output units. • Input and output units are connected through weighted interconnections. • Input and output vectors are perfectly correlated with each other component by
- 7. Maxnet • Symmetrical weights are present over the weighted interconnections. • Weights between neurons are inhibitory and fixed. • The maxnet with this structure can be used as a subnet to select a particular node whose net input is the largest. X1 Xm Xi Xj 1 1 1 −𝜀 −𝜀 −𝜀 −𝜀 −𝜀 −𝜀 1
- 8. Mexican Hat Net • Neurons are arranged in a linear order such that positive connections exist between Xi and neighborhood units & negative between Xi and far away units. • Positive region is Cooperation and negative region is Competition. • Size of these regions depend on the magnitude that exist between positive and X i X i+1 X i+2 X i+3 X i-1 X i-2 X i-3 W3 W3 W2 W2 W1 W1 𝛿𝑖 W0
Editor's Notes
- #3: Learning signal is the difference between the desired and actual response of a neuron. The perceptron learning rule is Consider a finite “n” number of input training vectors Associated target (desired) values x(n) and t(n) where n is from 1 to N Target is either +1 or -1 The output “y” is obtained on the basis of the net input calculated and activation function being applied over the net input