be reu and ai and human mindtheinput Dimension now see caras knows thatbefore this hidden layer there will bean input right there will be an inputlayer so I'm telling kasas that my inputlayer will have 32 neurons if you seeinput dim equals to this I telling Kazautomatically that it will have theinput layer will have 30 neurons so assoon as you do the first model. addright caras will go ahead and caras willDefine the input layer having 32 neuronsand the hidden layer number one whichwill have 16 neurons right so it is onlyin the first layer definition that weneed to Define how many input neurons weneed to have because it is the firsthidden layer that will be connected tothe input layer all right so after oncewhen we would have run this code rightyour architecture of the model willstart having the input layer and thehidden layer number one right and thesetwo will be connected by a denseconnection y16 y4 I have talked aboutthis previously as well that all ofthese are hyperparameters this is justan example that I giving you but whenyou will be actually running these deeplearning models you will have to testaround you will have to see your model'sperformance with different values of theneurons I have talked about thispreviously as well that the number ofneurons is ahyperparameter right this is just anexample that I'm giving you for theexample we need to have some valuesright so we have given16 renuka activation function is also ahyperparameter if you remember I haveasked you guys to keep on maintaining alist the number number of hidden layersthe number of neurons the activationfunction gradient descent all of thesewere hyperparameters this is just anexample I'm just explaining you how wecan build a neural network model but allof these values that you are seeing youwill have to decide them by trying outdifferent values all right now if yousee once when you have defined thehidden layer number one which will beconnected to the input layer right afterthat it's as simple as doing a model doadd if you want to add a new layer yougo ahead you do a model that do addright uh a dense layer I'm saying fourneurons activation is sigmoid right nowif you see I'm not telling Kaz that thislayer needs to be connected with theprevious layer Kaz automatically knowsthat whatever layers are coming afterthe definition of some layer right theywill be connected in a sequential mannerhow does it know that because it knowsthat this model object is of the typesequential right so it willautomatically keep on making theconnections as and when you keep ondefining the different layers right nowhere you see I have gone ahead and I'vehave added one one 2 3 4 5 6 7 8 ninelayers right so this model in total willhave 10 hidden layersnow you guys will be wondering I knowthat why do did we go ahead with 10hidden layers what is the answer with tothat layers is also a hyperparameterthis is just an example that I'm givingright this is just an example fourneurons 10 hidden layers activationfunction whatever you are seeing arejust hyperparameters how do we arrive atthese hyperparameters all of that willcome later right not justnow right now our objective is tounderstand how the K how can we buildthe Deep learning models using Kerasokay so don't worry about why I'm usingsigmoid why I'm using 4 16 and all Ihave explained this before as well theseare Hy hper parameters how do we arriveat the best value of the hyperparameters we will know that once whenwe are comfortable with defining andcreating these neural networkmodelsokay if you see if I go to the lastlayer the last layer that I have I'mdoing a model. add layers dods one andactivation I'm giving as send a soft Maxwhy do you think that I'm doing a denseof one over here can I go ahead and canI give two or three right so if you seeonly my the number of neurons of myoutput layer and the number of neuronsin the hidden layer are fixed all othersare totally upon your mercy right youcan go ahead and you can give any valuebut ideally you should give that valuewhich is giving you the best accuracy ofthe model that is hyperparameter tuningthat is for later all right 3 fourhidden layers right I'm just definingsome connections and then you have oneover here so your architecture iscomplete over here so finally you willbe getting all of these over hereokay I hope everyone is able tounderstand this piece of code the onethat I'vehighlighted so if I run this my modelwill be defined right the layers will bedefined I have a function known asmodel. summary this function will giveme how many parameters in total thismodel containsright all right if you look at thisfunction which is summary what thistells you is it tells you that how manynumber of neurons your hidden layernumber one has how many two has threehas four has till the last hidden layerand it also tells you the number ofparameters that this layer will have intotal right can you guys tell me how didwe arrive at this 528value mik so if you see right this willhappen over here in this partautomatically as soon as I say as soonas you say that the input Dimension is32 neurons caras will know that theinput layer which needs to be definedfirst will have 32 neurons after that wewe are defining the hidden layer numberone which will contain 16 neurons sothis statement this single statement isactually defining two layers for you itis first defining the hidden layer andthen it is defining the sorry it isfirst defining the input layer with 32neurons and then it is defining thehidden layer with 16 neurons exactly sothe rest are bias nodes right howbecause if youremember assume that youhave four inputs four neurons in thehidden layer assume that you have threesorry assume that you have four neuronsin the input layer and three neurons inthe hidden layer number one right howmany connections will these four andthese three have withinthem if we talk about the denseconnections I have shown

(Editor in charge: go)