Popular posts

Neural networks are modeled based on the functionality of human brain.our brain uses the interconnected network of neurons for processing/getting the information and model the world around us.Neurons get the information/input from other neurons using unit called dendrites.The neurons sums up all the information collected and if the resulting vales are greater than the threshold set it triggers.the triggered signal is then send to another neuron in the network through the axon.

Now lets come back to original question . what is neuron: neuron is the computation unit which takes single or multiple inputs and produce the output.The function which does the transformation from input to output can be simple logistic function or Sigmoid function, this unit unit is called activation function. Mathematical modelling for neuron:

the above fig shows a neurons connected with n other neurons thereafter receiving N inputs(I1,I2....In) this complete setup is called perceptron. each input will be associated with weight and both Input and weight can be real numbers. Perceptron = weights+Sum+f(Activation function) +bias each input is associated with the weight added together and passed into the activation function f. There are multiple types of Activation function i.e logistic,sigmoid, step function. the output of the step function will be 1 if input is higher than a certain threshold otherwise it will be 0. for example

I1=0.5 i1=1.0

weight w1=0.4 w2=0.9 Threshold =1.0

now sum = w1I1+w2i2=0.50.4+1.00.9 > 1.0 here total input is higher than the threshold thats why neuron fires.

Now the new question arrives how training happening in perceptron training happen is perceptron same as human brain learn , how human brains learn. how we teach child to recognize motor cycles? we show them examples like combination of motorcycles and teacher her this is motorcycle and this is not.

in our case input vectors from training sets are presented to perceptron one by one and weights are adjusted according to mathematical equation: for all input i:

Weight(i) =Weight(i)+b(T-O)P(i) Weight(i) is the weight vector P is the input vector O is the output given by perceptron T is the actual out for perceptron

when an entire pass through of all the input training vectors is completed without minimum error the perceptron training is completed.