Would you like to inspect the original subtitles? These are the user uploaded subtitles that are being translated: 1 00:00:00,560 --> 00:00:02,950 Arrive exciting tutorial ahead. 2 00:00:02,950 --> 00:00:04,850 Welcome back to the course on deep learning. 3 00:00:04,870 --> 00:00:08,090 Today we're talking about how neural networks work. 4 00:00:08,110 --> 00:00:13,840 Now we've led a lot of ground work we've talked about how neural networks are structured what elements 5 00:00:13,840 --> 00:00:16,420 they consist of and even their functionality. 6 00:00:16,660 --> 00:00:21,940 And today we're going to look at and a real example of how unusual neural network can be applied and 7 00:00:21,940 --> 00:00:27,760 we're actually going to work step by step through the process of its application so we know what is 8 00:00:27,760 --> 00:00:28,340 going on. 9 00:00:28,360 --> 00:00:30,690 So let's have a look what example we're going to be talking about. 10 00:00:30,790 --> 00:00:37,180 We're going to be looking at a property evaluation so we're going to look at a neural network that takes 11 00:00:37,180 --> 00:00:39,600 in some parameters of our property and value values. 12 00:00:39,640 --> 00:00:40,930 And the thing here. 13 00:00:40,960 --> 00:00:45,820 There's a small caveat for today's tutorial and that is we're not actually going to train the network. 14 00:00:45,820 --> 00:00:50,980 So a very important part in neural networks is training them up and we're going to look at that in the 15 00:00:50,980 --> 00:00:53,430 next tutorials in this section. 16 00:00:53,440 --> 00:00:57,760 For now we're going to focus on actual applications are we going to work with a neural network that 17 00:00:58,390 --> 00:01:04,870 we're going to pretend is already trained up and that will allow us to focus on the application side 18 00:01:04,870 --> 00:01:09,370 of things and not get bogged down in the training aspect and then we'll cover off the training when 19 00:01:09,370 --> 00:01:11,750 we already know the end goal we're working towards. 20 00:01:11,800 --> 00:01:12,640 Sounds good. 21 00:01:12,640 --> 00:01:14,900 All right let's jump straight into it. 22 00:01:15,250 --> 00:01:19,090 So let's say we have some input parameters. 23 00:01:19,090 --> 00:01:24,670 Right so let's say we have full parameters about the property we have area in square feet we have the 24 00:01:24,670 --> 00:01:29,950 number of bedrooms that distance the city and Miles the New York City and the age of the property and 25 00:01:29,950 --> 00:01:33,590 all of those four are going to comprise our inputs layer. 26 00:01:33,670 --> 00:01:40,810 Now of course they're probably way more parameters that define the price of a property but for simplicity 27 00:01:40,810 --> 00:01:43,000 sake we're going to look at just this for now. 28 00:01:43,010 --> 00:01:50,260 It's very basic form a neural network only has an input learn an output layer so no hidden layers and 29 00:01:50,260 --> 00:01:52,300 our output layer is the price that we're predicting. 30 00:01:52,300 --> 00:02:00,990 So in this form what these inputs variables would do is they would just be weighted up by the synopses 31 00:02:01,030 --> 00:02:04,150 and then the output there would be calculated. 32 00:02:04,150 --> 00:02:06,190 Basically the price would be calculated and would get a price. 33 00:02:06,190 --> 00:02:14,250 And for instance the price could be calculated as simple as the weighted sum of all of the inputs. 34 00:02:14,290 --> 00:02:17,720 And again here you could use pretty much any function you could use. 35 00:02:17,830 --> 00:02:25,330 What we're using now we could use any of the activation functions we had previously you could use logistic 36 00:02:25,330 --> 00:02:26,020 regression. 37 00:02:26,050 --> 00:02:32,680 You could use a squared function you can do pretty much anything here but the point is that you get 38 00:02:32,740 --> 00:02:33,570 a certain output. 39 00:02:33,600 --> 00:02:40,980 And moreover most of the machine learning algorithms that exist can be represented in this form and 40 00:02:40,990 --> 00:02:46,390 this is basically a diagrammatic representation of how you deal with the variables or by changing the 41 00:02:46,390 --> 00:02:51,910 way it's a formalised you can accomplish quite a lot of the machine learning algorithms that we've talked 42 00:02:51,910 --> 00:02:58,510 about before and put them into this form and that just tends to show how powerful Noul are neural networks 43 00:02:58,510 --> 00:02:58,900 are. 44 00:02:58,960 --> 00:03:04,270 Even without the hidden layers we are ready where we have a representation that works for most other 45 00:03:04,270 --> 00:03:05,360 machine learning algorithms. 46 00:03:05,470 --> 00:03:13,210 But in neural networks what we do have is an advantage that gives us lots of flexibility and power which 47 00:03:13,600 --> 00:03:17,050 is where that increase in accuracy comes from. 48 00:03:17,050 --> 00:03:24,490 And that power is the hidden layers and there we go that's our hit Alair we added it in and now we're 49 00:03:24,490 --> 00:03:30,100 going to understand how that hidden lair gives us that extra power. 50 00:03:30,300 --> 00:03:35,830 And in fact to do that we're going to walk through an example so as we agreed this neural network has 51 00:03:35,830 --> 00:03:40,150 really been trained up and now we're just going to plug in we're going to imagine they were plugging 52 00:03:40,150 --> 00:03:48,370 in a property and we're going to walk step by step through how the neural network will deal with the 53 00:03:48,370 --> 00:03:51,880 input variables and calculate the Hindol area and then calculate the output. 54 00:03:51,880 --> 00:03:54,430 So let's go through this is going to be exciting. 55 00:03:54,430 --> 00:03:54,880 All right. 56 00:03:55,060 --> 00:04:00,950 We've got all four variables on the left and we're going to first start with the top Nurin on the Hindle 57 00:04:00,980 --> 00:04:01,420 there. 58 00:04:01,450 --> 00:04:08,290 Now as we previously saw in the press literals all of the neurons from the input layer they have Cynapsus 59 00:04:08,290 --> 00:04:12,940 connecting it to each one of them to the top neuron in the hidden lair. 60 00:04:13,480 --> 00:04:15,240 And those systems have weights. 61 00:04:15,250 --> 00:04:20,470 Now let's agree that some weights will have a non-zero value some ways will have zero value because 62 00:04:20,710 --> 00:04:28,390 basically not all inputs will be valid or not all inputs will be important for every single neuron sometimes 63 00:04:28,390 --> 00:04:29,420 inputs will not be important. 64 00:04:29,420 --> 00:04:34,180 Here we can see two examples that X-1 next three the area and the distance to the city and Miles are 65 00:04:34,180 --> 00:04:38,770 important for that neuron whereas bedrooms and age are not like let's think about this for a second 66 00:04:38,770 --> 00:04:40,540 why how would that be the case. 67 00:04:40,550 --> 00:04:47,350 Like why would a neuron be linked to the area and the distance what does that what could that mean. 68 00:04:47,350 --> 00:04:53,410 Well that could mean that normally the further away you get from the city the cheaper real estate becomes 69 00:04:53,410 --> 00:04:58,420 and therefore the space in square feet of properties becomes larger. 70 00:04:58,420 --> 00:05:02,380 So for the same price you can get a larger property the further away you go from the city that's normal 71 00:05:02,380 --> 00:05:02,640 right. 72 00:05:02,640 --> 00:05:08,930 That that makes sense and probably what this neuron is doing is it is looking specifically it's like 73 00:05:08,950 --> 00:05:16,990 like a sniper it's looking for area properties which have which are not so far from the city but have 74 00:05:16,990 --> 00:05:17,630 a large area. 75 00:05:17,630 --> 00:05:24,830 So for their distance from the city they have an unfair square foot area. 76 00:05:24,850 --> 00:05:28,690 Right so something as abnormals height is higher than average so they're quite close to the city but 77 00:05:28,690 --> 00:05:35,290 they're still large as opposed to the other ones at the same distance and so that neuron again we're 78 00:05:35,290 --> 00:05:40,450 speculating here but that neuron might be picking up laser picking out those specific properties and 79 00:05:40,450 --> 00:05:46,150 it will activate and hence the activation function it will activate it'll fire up only when the certain 80 00:05:46,150 --> 00:05:50,830 criteria is met that you know the distance and the area of the proper distance to Syrian air of the 81 00:05:50,830 --> 00:05:57,580 area of the property and it performs on calculations inside itself and it combines those two and as 82 00:05:57,580 --> 00:06:01,990 soon as certain areas where it fires up and that contributes to the price in output. 83 00:06:02,260 --> 00:06:06,760 And therefore this neuron doesn't really care about bedrooms and age of the property because it's focused 84 00:06:06,760 --> 00:06:07,730 on that specific thing. 85 00:06:07,750 --> 00:06:12,640 That's where the power of the neural network comes from because you have many of these years and will 86 00:06:12,640 --> 00:06:14,390 see just now how the other ones work. 87 00:06:14,650 --> 00:06:21,370 But what I want to agree here is that let's not even draw these lines for the synopses that are not 88 00:06:22,090 --> 00:06:26,170 in place so that we don't clutter up our image as the only reason we're not going to draw them so let's 89 00:06:26,170 --> 00:06:27,290 just get rid of those too. 90 00:06:27,460 --> 00:06:33,940 And that way we will know exactly OK so this neuron is focused on area and distance to the city. 91 00:06:33,940 --> 00:06:34,520 All right. 92 00:06:34,900 --> 00:06:35,980 So as always we agree on that. 93 00:06:35,980 --> 00:06:36,830 Let's move on to next. 94 00:06:36,850 --> 00:06:42,400 Let's take them one in the middle here we've got three parameters feeding into this neuron so we've 95 00:06:42,400 --> 00:06:46,010 got the area the bedrooms and the age of the property. 96 00:06:46,030 --> 00:06:48,600 So what could be the reason here. 97 00:06:48,600 --> 00:06:55,660 Let's again let's try to understand the intuition and the thinking of this neuron how is this neuron 98 00:06:55,660 --> 00:06:56,110 thinking. 99 00:06:56,110 --> 00:06:57,770 Why is it picking these two parents. 100 00:06:57,780 --> 00:07:02,110 What could it be what could have a hit like found in the data. 101 00:07:02,110 --> 00:07:06,580 Right so we've already established this trained up data set the training has happened a long time ago 102 00:07:06,580 --> 00:07:11,590 maybe like a day ago or somebody is written up as it is now we're just applying and we know that this 103 00:07:11,590 --> 00:07:17,650 neuron through all of the thousands of examples of properties has found out that the area plus the bedrooms 104 00:07:17,650 --> 00:07:20,590 plus the age combination of those parameters is important. 105 00:07:20,590 --> 00:07:21,600 Why could that be the case. 106 00:07:21,610 --> 00:07:29,410 Well for instance maybe in that specific city in those suburbs that this neural network has been trained 107 00:07:29,410 --> 00:07:38,860 up in perhaps there's a lot of families with kids who have two or more children who are looking for 108 00:07:40,000 --> 00:07:47,470 large properties with lots of bedrooms but which are new rights which are not old proper because maybe 109 00:07:47,590 --> 00:07:52,910 that's in that area almost appropriate or kind of like big properties are usually old. 110 00:07:52,930 --> 00:07:59,860 But there's lots of modern families and maybe there has been a social demographic shift and or maybe 111 00:07:59,860 --> 00:08:07,540 there's been like a lot of like some growth in terms of employment and jobs for the younger self population 112 00:08:07,570 --> 00:08:15,520 maybe just you know the like the population demographics have changed and now younger couples or younger 113 00:08:15,520 --> 00:08:22,210 families are looking for properties but they prefer new properties so they want the age of the property 114 00:08:22,210 --> 00:08:27,370 to be lower and hence from the training that this neural network has undergone. 115 00:08:27,370 --> 00:08:32,890 It knows that when there's a property with a large area and with lots of bedroom with these three at 116 00:08:32,890 --> 00:08:36,790 least three bedrooms for the parents for the first of the second child for at least three bedrooms maybe 117 00:08:36,790 --> 00:08:45,190 a guest room when you property with high area and lots of bedrooms that is valued that in that market 118 00:08:45,190 --> 00:08:48,270 that is valuable so that Meuron has picked that up. 119 00:08:48,280 --> 00:08:49,390 It knows that. 120 00:08:49,420 --> 00:08:51,490 OK so this is what I'm going to be looking for. 121 00:08:51,490 --> 00:08:56,620 I don't care about the distance to the city and Miles wherever it is as long as it has high area lots 122 00:08:56,620 --> 00:08:57,080 of bedrooms. 123 00:08:57,110 --> 00:09:01,750 As soon as that criteria is met the neuron fires up and the combination of these two parameters and 124 00:09:01,750 --> 00:09:06,190 this is again this is where the power of the neural network is coming from because it combines these 125 00:09:06,190 --> 00:09:11,300 two parameters into a brand new parameter into brand new attributes. 126 00:09:11,470 --> 00:09:17,860 That helps with the evaluation that helps with the valuation of the property. 127 00:09:17,860 --> 00:09:21,910 It combines them into a new attribute and therefore it's more precise. 128 00:09:21,910 --> 00:09:24,160 So there we go that's how that works. 129 00:09:24,250 --> 00:09:30,040 And let's look at another one let's look at the very bottom one for instance this neuron could be could 130 00:09:30,040 --> 00:09:35,110 even have picked up just one pair and that it could have just picked up eight and not in any of the 131 00:09:35,110 --> 00:09:35,540 other ones. 132 00:09:35,540 --> 00:09:37,530 And how could that be the case. 133 00:09:37,900 --> 00:09:45,760 Well this is a classic example of when age could mean like as we all know the older properties usually 134 00:09:45,760 --> 00:09:47,710 it's less valuable because it's worn out. 135 00:09:47,710 --> 00:09:50,740 Probably the building is old probably you know things are falling apart. 136 00:09:50,740 --> 00:09:55,460 More maintenance is required so the price drops in terms of the price of the real estate. 137 00:09:55,480 --> 00:09:58,880 Whereas a brand new building it would be more expensive because it's brand new. 138 00:09:59,200 --> 00:10:03,980 Perhaps if a property is over a certain age that could indicate that it's a historic property. 139 00:10:04,030 --> 00:10:11,240 For instance if a property is under 100 years old then the older it is the less valuable it is. 140 00:10:11,380 --> 00:10:17,050 But as soon as it jumps over 100 years old all of a sudden it becomes a historic property because this 141 00:10:17,050 --> 00:10:20,840 is a property where people still have hundreds of years ago. 142 00:10:20,860 --> 00:10:26,240 It tells a story it's got all this history behind it and some people like that some people value that. 143 00:10:26,290 --> 00:10:32,110 In fact quite a lot of people would like that and would be proud to live in a property and especially 144 00:10:32,110 --> 00:10:36,970 in the higher socioeconomic classes they would they would show off to their friends or things like that 145 00:10:37,000 --> 00:10:42,330 and therefore properties that are over 100 years old could be deemed as historic. 146 00:10:42,330 --> 00:10:47,890 And therefore this neuron as soon as it sees a property over 100 years old it'll fire up and contribute 147 00:10:48,430 --> 00:10:49,360 to the overall price. 148 00:10:49,360 --> 00:10:55,090 And otherwise if it's under 100 years old then it won't work and this is a good example of that. 149 00:10:55,180 --> 00:10:57,580 The rectifier function being applied. 150 00:10:57,580 --> 00:11:04,630 So here you've got like a very like a zero until a certain point and then let's say 100 years old and 151 00:11:04,630 --> 00:11:09,340 then after 100 years old the older it gets the higher the value the higher the contribution of this 152 00:11:09,340 --> 00:11:11,310 neuron to the overall price. 153 00:11:11,470 --> 00:11:18,910 And there's just a wonderful example of a very simple example of this rectifier function in action. 154 00:11:18,910 --> 00:11:19,720 So there we go. 155 00:11:19,750 --> 00:11:20,960 That could be this year. 156 00:11:21,100 --> 00:11:27,840 And moreover the neural network could have picked up things that we wouldn't have thought of ourselves 157 00:11:27,850 --> 00:11:28,530 right. 158 00:11:28,570 --> 00:11:34,360 For instance bedrooms plus distance the city maybe that's in combination somehow contributes to the 159 00:11:34,360 --> 00:11:38,710 price maybe not as strong as the other neurons and it contributes but it still contributes or maybe 160 00:11:38,890 --> 00:11:43,510 it detracts from the price that could also be the case or other things like that and maybe add your 161 00:11:43,510 --> 00:11:51,040 own picked up all for a combination of all four of these parameters and as you can see that these neurons 162 00:11:51,040 --> 00:11:58,840 this whole hidden layer situation allows you to increase the flexibility of your neural network and 163 00:11:58,840 --> 00:12:05,890 allows you to really allows the neural network to look for very specific things and then in combination 164 00:12:06,070 --> 00:12:08,260 that's where the power comes from. 165 00:12:08,260 --> 00:12:12,570 It's like that example the answer I'd like an ad by itself cannot build an anthill. 166 00:12:12,580 --> 00:12:17,920 But when you have like a thousand or 100000 ads they can build an anthill together and that's that's 167 00:12:17,920 --> 00:12:18,600 the situation here. 168 00:12:18,610 --> 00:12:20,990 Each one of these neurons by itself cannot predict the price. 169 00:12:21,100 --> 00:12:27,910 But together they have super powers and they predict the price and they can do quite an accurate job 170 00:12:27,940 --> 00:12:30,300 if trained properly set up properly. 171 00:12:30,730 --> 00:12:35,410 And that's what this whole Course is about understanding how to utilize them. 172 00:12:35,410 --> 00:12:42,430 There we go so that is a step by step example and walkthrough of how neural networks actually work. 173 00:12:42,460 --> 00:12:45,590 I hope you enjoyed today's tutorial and I can't wait to see you next time. 174 00:12:45,610 --> 00:12:47,350 Until then enjoy learning. 20265