Would you like to inspect the original subtitles? These are the user uploaded subtitles that are being translated: 1 00:00:01,130 --> 00:00:03,280 Step two is basically the same thing. 2 00:00:03,290 --> 00:00:08,900 But now the networks have been more trained which is going to do this again to reiterate this whole 3 00:00:08,900 --> 00:00:11,420 process and understand it better. 4 00:00:11,420 --> 00:00:18,450 So Step Two noise goes into the generator generator generates images which are not. 5 00:00:18,740 --> 00:00:20,570 They don't look as random anymore. 6 00:00:20,570 --> 00:00:28,230 They're kind of a bit clearer and we'll understand why in just a second off once we were finishing up 7 00:00:28,230 --> 00:00:31,580 with Step two we'll understand how they generate or understand these things. 8 00:00:31,670 --> 00:00:38,030 But basically through the through the back propagation error the declaration of error understood where 9 00:00:38,020 --> 00:00:40,730 it was making mistakes in the it. 10 00:00:41,120 --> 00:00:46,320 It's adjusted its weight and now is generating images which are a little bit more like dogs. 11 00:00:47,760 --> 00:00:53,270 And now we're going to train the discriminator again and for that we need some images of dogs. 12 00:00:53,490 --> 00:00:54,410 A new batch. 13 00:00:54,410 --> 00:01:01,390 And now we're going to put all of those into the discriminator along with the images of the dogs and 14 00:01:02,140 --> 00:01:07,900 it will output some values so I'll put the values now we need to compare the values to the actual numbers 15 00:01:08,210 --> 00:01:13,420 that we want that we know which ones are those which are not dogs the top ones are not dogs but the 16 00:01:13,420 --> 00:01:14,110 dogs. 17 00:01:14,290 --> 00:01:21,940 So we tell that to the discriminator this cremator Kalka is the era that propagates the arrow through 18 00:01:21,940 --> 00:01:26,650 its network and therefore learns from that and the weights are obtained. 19 00:01:26,650 --> 00:01:30,870 So next time we'll do a better job at discriminating dogs versus dogs. 20 00:01:30,940 --> 00:01:31,810 So it's learning. 21 00:01:31,810 --> 00:01:37,120 So it's now at some level to kind of like metaphorically speaking. 22 00:01:37,660 --> 00:01:41,870 And now we want to train the generators so the. 23 00:01:42,070 --> 00:01:46,470 So we want to get rid of that the generator generator and then we're going to use the same image. 24 00:01:46,480 --> 00:01:50,790 We're going to put them through the network of the discriminator is going to output some values. 25 00:01:50,830 --> 00:01:52,710 As you can see these values are lower. 26 00:01:52,750 --> 00:01:58,840 So if I go back you'll see that the values were apparently is open for the point to now the values are 27 00:01:58,990 --> 00:02:07,360 0.5 inches or 21 so they draw because the discriminator has learned what how that these don't really 28 00:02:07,360 --> 00:02:08,010 look like dogs. 29 00:02:08,020 --> 00:02:09,630 But as you can see they don't. 30 00:02:09,630 --> 00:02:13,780 They didn't drop as low they were like dropping to like a complete zero. 31 00:02:13,790 --> 00:02:18,040 We'll talk about that later further down when we're doing Step Three. 32 00:02:18,070 --> 00:02:24,390 But for now just kind of noticed that what do we want to discuss now though is that what happens. 33 00:02:24,400 --> 00:02:27,390 How does a generator kind of like get better. 34 00:02:27,550 --> 00:02:29,860 Well I think of that intuitively. 35 00:02:29,950 --> 00:02:35,000 Well we again we take these values we calculate errors because we want these values to be ones. 36 00:02:35,110 --> 00:02:40,510 That's what the generator wanted wants to trick the discriminator and this error these errors of backburn 37 00:02:40,510 --> 00:02:45,340 brigade through to the discriminator and the weights are updated and the way to think of it intuitively 38 00:02:45,340 --> 00:02:55,390 is these networks are huge What's what we have drawn here is like these are very small just representations 39 00:02:55,390 --> 00:02:58,950 of those networks just like images to show that this is indeed a network. 40 00:02:58,960 --> 00:03:01,250 But in reality those areas are much much bigger. 41 00:03:01,480 --> 00:03:08,230 And the way to think about it is that the there's the communication happening between the general discriminant 42 00:03:08,260 --> 00:03:11,260 So the generator creates these images. 43 00:03:11,280 --> 00:03:14,580 Since this criminal says hey this computer I have some images here. 44 00:03:14,710 --> 00:03:17,350 Do you do you think these are images of dogs or not. 45 00:03:17,350 --> 00:03:18,800 And what do you think the possibilities are. 46 00:03:18,970 --> 00:03:23,790 And the distributor looks at them and says oh well you know those don't really look like dogs to me 47 00:03:23,800 --> 00:03:28,270 I'll give them about 20 percent probability 10 percent probability of 50 percent possibility and then 48 00:03:28,270 --> 00:03:30,550 that discriminators like Are you OK. 49 00:03:30,580 --> 00:03:31,690 You're totally right. 50 00:03:31,720 --> 00:03:32,770 I was trying to trick you. 51 00:03:32,770 --> 00:03:35,700 These are not dogs but can you tell me what I did wrong. 52 00:03:35,710 --> 00:03:38,670 You know like what where did I go wrong. 53 00:03:38,920 --> 00:03:45,070 And the discriminator in this case my look at the images and say this is the back propagation we're 54 00:03:45,070 --> 00:03:51,700 kind of like this is the intuitive understanding of the back propagation process and gry indecent the 55 00:03:53,490 --> 00:03:59,440 discriminator might say something like look I checked your images of dogs you know you know the images 56 00:03:59,440 --> 00:04:02,860 you send me and I really have seen images of dogs. 57 00:04:03,370 --> 00:04:06,910 So I know that dogs normally for instance have eyes. 58 00:04:06,910 --> 00:04:13,240 Your dogs have some little black dots which don't really look like eyes to me and you're missing out 59 00:04:13,240 --> 00:04:17,680 on eyes or you know you're you know we don't have enough paws or you don't have tails and your dogs 60 00:04:17,680 --> 00:04:19,480 those don't look like dogs to me or they're not. 61 00:04:19,690 --> 00:04:24,580 They don't look three dimensional enough they look very flat dogs and all three look three dimensional 62 00:04:24,580 --> 00:04:25,130 to me. 63 00:04:25,150 --> 00:04:28,490 So and that's this process of back propagation. 64 00:04:28,540 --> 00:04:35,230 This error is actually that's what it's kind of like overall in human language telling that discriminate 65 00:04:35,410 --> 00:04:42,530 the generator but in reality of course it's just telling the generator that hey you know you got to 66 00:04:42,740 --> 00:04:45,970 go to this part of your image doesn't feel right. 67 00:04:45,970 --> 00:04:50,130 This part of the image doesn't feel right and just basically dealing with individual pixels and how 68 00:04:50,130 --> 00:04:54,790 it's how they should be updated for these images to better resemble dogs. 69 00:04:54,790 --> 00:04:57,750 And then the weights of the neural network are up. 70 00:04:58,090 --> 00:05:00,440 So the next time we'll do a better job the generator. 71 00:05:00,640 --> 00:05:01,390 So step three. 7884