Would you like to inspect the original subtitles? These are the user uploaded subtitles that are being translated: 1 00:00:00,360 --> 00:00:06,600 Hello and welcome to this new tutorial in the previous Statoil we tackled the first two steps steps 2 00:00:06,690 --> 00:00:09,090 of the training of the discriminator. 3 00:00:09,180 --> 00:00:13,830 We trained it on real images and then on fake images. 4 00:00:13,830 --> 00:00:15,500 And so now it's almost over. 5 00:00:15,630 --> 00:00:23,850 We are ready for the third substate which is to get the total air as the sum of the to the R R D real 6 00:00:23,940 --> 00:00:25,380 and b r d fake. 7 00:00:25,530 --> 00:00:31,260 So we'll get this some that will get as little error and then we'll back propagate the stall error back 8 00:00:31,260 --> 00:00:36,050 into the new one that work at the discriminator to then update the weights through to get to Graylands 9 00:00:36,050 --> 00:00:39,430 descent according to how much they're responsible for the error. 10 00:00:39,660 --> 00:00:40,650 So very easy now. 11 00:00:40,650 --> 00:00:43,330 Let's start by getting the total error. 12 00:00:43,350 --> 00:00:46,160 We're going to call it the R R D. 13 00:00:46,200 --> 00:00:49,140 It's the total error of the discriminator. 14 00:00:49,200 --> 00:01:02,940 So e r d equals well R R D real which I am copying and pasting Plus R D fake again which I'm copying 15 00:01:03,060 --> 00:01:04,670 and pasting. 16 00:01:04,710 --> 00:01:06,760 And there we go we have all error. 17 00:01:06,930 --> 00:01:12,840 So now let's back propagate it back into the new one that work of the discriminator and to do this. 18 00:01:12,840 --> 00:01:14,050 Thanks Supai torch. 19 00:01:14,250 --> 00:01:20,850 It's really simple we just need to take the error the total error the R D and then dot and then we use 20 00:01:20,850 --> 00:01:25,250 the back word function to back propagate it. 21 00:01:25,470 --> 00:01:26,260 Perfect. 22 00:01:26,430 --> 00:01:33,180 And now one final step of this first big step of the training we need to apply is to cast a grade in 23 00:01:33,180 --> 00:01:37,660 the center of the weight and to do this guess what we're going to take right now. 24 00:01:37,660 --> 00:01:41,480 We're going to take the optimizer of the discriminator. 25 00:01:41,480 --> 00:01:48,170 And so I'm copying this I'm pasting it here and there we go almost over. 26 00:01:48,210 --> 00:01:57,030 We just need to add a dirt and apply the step function to step function applies the optimizer on the 27 00:01:57,030 --> 00:02:03,110 neural network to discriminator to have they the weight of the discriminator according to how much they're 28 00:02:03,120 --> 00:02:09,000 responsible for the total loss error which is the sum of the real error and the fake error. 29 00:02:09,210 --> 00:02:10,410 And so now well done. 30 00:02:10,500 --> 00:02:11,230 Good job. 31 00:02:11,280 --> 00:02:17,370 You're ready to move on to the second step of the training and we'll take care of that in the next tutorial. 32 00:02:17,370 --> 00:02:19,170 Until then enjoy computer vision. 3294