WEBVTT 00:02.070 --> 00:04.370 Hello and welcome to this tutorial. 00:04.620 --> 00:06.050 In previous Teito DLP. 00:06.110 --> 00:12.420 How understood input layer and convolution layer in this little deal, we will understand the remaining 00:12.420 --> 00:14.800 layers in convolutional neural network. 00:15.180 --> 00:16.110 Let us begin. 00:19.230 --> 00:23.220 Third layer is Relu Layer and Elu stands for Rectified Línea. 00:23.250 --> 00:23.760 Do you need? 00:24.000 --> 00:26.670 It is widely used activation function. 00:28.050 --> 00:31.980 This is the mathematical equation for RELU activation function. 00:33.390 --> 00:35.220 Output of this layer is podgy. 00:35.220 --> 00:38.330 To add value ranges from zero to infinity. 00:40.090 --> 00:43.510 Nature of this activation function, Relu is non-linear. 00:44.050 --> 00:51.040 We can easily back propo get errors and multiple layers of neurons being activated by the RELU function. 00:52.570 --> 00:59.200 The RELU activation function is computationally less expensive than the other activation functions. 00:59.740 --> 01:07.390 And this is because at the time, only few neurons are activated to add non-linearity to our model. 01:07.570 --> 01:08.890 We use this layer. 01:12.040 --> 01:19.720 Pooling layer deep pooling layer is responsible for reducing this vital signs of Convolve features. 01:21.310 --> 01:27.470 We add a pooling layer to decrease the computational power that is required to process the data. 01:29.200 --> 01:35.240 The pulling layer is useful for extracting the dominant features, which are rotational and positional 01:35.240 --> 01:38.690 invariant so we can play in do more efficiently. 01:39.350 --> 01:45.200 In simple words, we can say that we add puling layer to reduce the size of images. 01:48.390 --> 01:55.830 There are two types of pulling, Max, bullying and average pulling max pulling returns, the maximum 01:55.830 --> 01:59.910 value from a portion of inmates that is covered by colonel. 02:01.470 --> 02:08.250 And the average pulling returns, the average of all the values from a portion of inmate that is covered 02:08.280 --> 02:08.790 by a. 02:12.660 --> 02:17.520 Let us understand the mathematical calculation of Max pulling and average pulling. 02:18.120 --> 02:24.960 First, we will understand Max pulling in, Max pulling this Telek, the maximum value here in this 02:24.960 --> 02:25.710 section. 02:26.070 --> 02:28.830 Twenty is the maximum value here. 02:28.860 --> 02:32.430 Thirty thirty here in this section. 02:32.670 --> 02:36.060 Maximum value is 112, 112. 02:36.660 --> 02:42.530 And here in this last section, the maximum value is 37 37. 02:43.350 --> 02:44.220 Let us understand. 02:44.340 --> 02:49.440 Average pulling 13 is the average of these four values. 02:50.280 --> 02:54.480 Similarly, it is the average of these four values. 02:55.020 --> 02:58.650 Seventy nine is the average of these four values. 02:59.040 --> 03:02.340 And 20 is the average of these four values. 03:04.140 --> 03:07.710 So this is home, Max, pulling an average pulling calculated. 03:10.790 --> 03:15.510 Let us understand the fully connected layer in fully connected layer. 03:15.730 --> 03:22.730 They passed the output of pulling layer, so input of deeply connected layer is the output of pulling 03:22.730 --> 03:25.550 layer and after the input layer. 03:25.580 --> 03:32.210 There are multiple layers in fully connected layer, and these multiple layers are the hidden layers. 03:32.810 --> 03:36.680 And at the end, we get output from that fully connected layer. 03:37.460 --> 03:41.870 You can think this fully connected layer is artificial neural network. 03:44.860 --> 03:50.340 So finally, this is the complete architecture of convolutional neural network flight. 03:50.590 --> 03:52.300 We have to provide the input. 03:52.900 --> 03:55.840 Then we have to add convolution and relu layer. 03:56.080 --> 03:59.140 And after that, we have to add a pooling layer. 03:59.770 --> 04:04.570 There can be multiple pairs of convolution plus Trelew and pooling layers. 04:05.230 --> 04:07.900 After that, there is fully connected layer. 04:08.560 --> 04:12.430 We will get our output at the end of fully connected layers. 04:13.510 --> 04:17.420 So dictatorial about the convolutional neural network NCEA. 04:17.800 --> 04:19.660 I will see you in the next one. 04:19.900 --> 04:21.730 buildOn happy learning.