00:00glance it through once more so we have this input features x1 x2 to xn and the perceptron
00:10it receives n features as input x1 x2 to xn and these features are connected with some weights
00:20which once again i repeat that these weights decide how much of importance we give to each
00:27and every input and once they come and get connected here in this neuron there are two
00:36activities that are done there one is the summation which is basically the multiplication
00:44of the weight with the input it is a linear combination of all the weights into input
00:51and then the output will pass through some mathematical function which we call as the
00:58activation function and then that result from the activation function goes as the output we will
01:05have much more clarity of how this calculations happen in multilayer perceptron when we go to the
01:13next slide now one thing i would like to highlight here is in if you want a perceptron to be
01:20working
01:20the input features must be numeric that means some numbers if we have some non-numeric input features
01:29they needs to be converted to numeric ones in order to use a perceptron so these are the requirements
01:35for working of a perceptron model okay then we go and see what is a feed forward neural network
01:47and this feed forward neural network is basically the way a multilayer perceptron works so we will see
01:56we have an input layer that we know and in this input layer we have two input neurons x1 and
02:02x2
02:04then we have two neurons in the hidden layer now we will see how these connections how these weights
02:12and other things are you know given in a multilayer perceptron model
02:20you see the weight associated with the x1 input neuron is w11 and weight associated with the neuron x2
02:31the input neuron x2 is w21 just keep an eye on the nomenclature how we are naming these weights
02:40it is from where it is going to whichever neuron it is going so w11 refers to the weight this
02:47is the
02:47nomenclature we will be using in our lecture w11 means it is going from x1 to the first neuron in
02:55the hidden layer w21 means it is going from x2 to the first neuron in the hidden layer similarly x1
03:05is
03:05connected to the second neuron in the hidden layer and the associated weight is w12 and even x2 is
03:12connected to the hidden layer neuron here it is the weight associated is w22 then there is a term
03:21which is the bias term which is which goes to both the neurons of the hidden layer now what happens
03:31after these connections are done inside the hidden layer as i told in the last slide there are two
03:38mathematical operations that happens one is the summation and another one is the activation function
03:44we will see exactly what happens in those two steps so the first part it is a weighted sum of
03:53the inputs
03:53what do i mean by the weighted sum of the inputs it means the first part takes the weight w11
04:01multiplied with the input neurons value x1 plus w21 which is the weight associated with x2 when it is
04:11connected to the first neuron in the hidden layer w21 into x2 plus the bias term b1 similarly for the
04:24second neuron in the hidden layer the weighted sum of inputs is said to equal to w12 which is what
04:31is w12 here this is w12 is the weight associated with x1 input when it is going to your second
04:39neuron
04:40in the hidden layer so w12 into x1 plus w22 into x2 plus b1 so this is called as the
04:47weighted sum of
04:48the inputs because the inputs we are multiplying with the weights to give a certain amount of importance
04:55to the input neurons okay and the biases we are adding so this is the first part of the calculation
05:03then what happens after this after this there is an activation function we have not discussed
05:10activation function till now it will come in my next session but for time being we will understand
05:16that activation function is a mathematical function which takes this weighted sum of the input passes
05:23through the mathematical function to give an output which goes to the neuron in the next layer so here
05:32the activation function is g which takes z1 passes z1 through the activation function later we will
05:40discuss what type of activation function how they change and other things right now it is only a
05:45mathematical function operation and then it gives the output a1 similarly in the neuron 2 of the hidden layer
05:52also the activation function is g which takes input z2 which is my weighted sum of inputs
06:01and then it gives the output a2 now this is the work that is happening inside the hidden layer then
06:09we have
06:11another neuron which is basically the output layer and this neuron gets the input from neuron 1 of the
06:21hidden layer and neuron 2 of the hidden layer now as i told after the activation function is working on
06:26z1
06:26we get a1 as the output and this a1 is now going to the output neuron and once again we
06:35have the weighted sum
06:36of the inputs z3 equal to w1k which is the weight associated with this connection into a1 plus w2k the
06:46weight
06:47associated with the connection between the second neuron of the hidden layer and the output neuron
06:52and we get the weighted sum of the inputs in this case there is no bias which is taken here
06:59future nevertheless
06:59now
06:59now that
06:59now that
Comments