1 00:00:01,500 --> 00:00:08,550 Here we have two systems, one is artificial neural network, which we want to design, the other one 2 00:00:08,550 --> 00:00:09,800 is just the process. 3 00:00:10,320 --> 00:00:11,960 It can be any process. 4 00:00:11,970 --> 00:00:15,050 It can be the correlation between to a stock market. 5 00:00:15,450 --> 00:00:22,880 It can be a doctor who wants to make some decision for a patient or any other processes. 6 00:00:24,210 --> 00:00:25,740 Here we have our input. 7 00:00:25,740 --> 00:00:29,790 We just feed the input into both systems. 8 00:00:30,120 --> 00:00:37,650 And let's see what would be the output of an actual system, a real system and a system which we designed 9 00:00:37,650 --> 00:00:39,640 using artificial neural network. 10 00:00:40,440 --> 00:00:47,490 For this one here, we have a function of X, which is of our input based on our teta, which is of 11 00:00:47,490 --> 00:00:48,420 our parameters. 12 00:00:48,690 --> 00:00:56,490 We already discussed this parameter are the weight and Bias's number of layers which we can adjust. 13 00:00:57,000 --> 00:00:59,430 Suppose here the output is why. 14 00:00:59,970 --> 00:01:02,460 Let's call the other output y hat. 15 00:01:02,940 --> 00:01:08,130 The best system is when Y is equal to y hat. 16 00:01:09,210 --> 00:01:15,930 If we are looking actually for this result, we want to have a system, an artificial neural network, 17 00:01:15,930 --> 00:01:20,990 which gives us the same output as other process. 18 00:01:21,510 --> 00:01:25,220 But is it actually it's actually possible. 19 00:01:25,320 --> 00:01:32,790 Let's check it out to you and describe here I have I and my hand. 20 00:01:32,790 --> 00:01:38,120 If we plot them then all the data should be exactly in this line. 21 00:01:38,430 --> 00:01:42,280 It means one is equal to Y and this is the best system. 22 00:01:42,300 --> 00:01:45,180 This is ideal system that we are looking for. 23 00:01:45,930 --> 00:01:47,030 In reality. 24 00:01:47,580 --> 00:01:49,250 Usually this won't happen. 25 00:01:50,130 --> 00:01:51,450 There are several reasons. 26 00:01:51,450 --> 00:02:00,840 For example, if we have a natural system, usually it's nearly impossible to include all the factors 27 00:02:00,840 --> 00:02:03,410 which have effect on our output. 28 00:02:04,080 --> 00:02:09,280 For example, in moon can effect on our health. 29 00:02:09,750 --> 00:02:18,230 So this is a parameter which doctors usually don't use it when they want to make some decision for their 30 00:02:18,240 --> 00:02:19,010 patients. 31 00:02:19,380 --> 00:02:27,360 But what if this factor is very important that can change the output, but we really don't know how 32 00:02:27,360 --> 00:02:34,080 to use it in how to measure it so we don't include it in our artificial neural network. 33 00:02:34,410 --> 00:02:42,000 So especially for the natural systems, it's nearly impossible to have the same output for Y and Y, 34 00:02:42,000 --> 00:02:46,010 and that's why we are looking for minimizing this error. 35 00:02:48,440 --> 00:02:54,810 If we subtract the output of Y from Y hat, then we should have our error. 36 00:02:55,340 --> 00:02:58,650 So this is the error that we are looking for minimizing. 37 00:02:59,030 --> 00:03:00,610 We are looking for minimizing. 38 00:03:00,620 --> 00:03:08,180 Usually we use a or it can be minimizing the absolute value of it. 39 00:03:09,320 --> 00:03:18,980 If it goes to zero, if E goes to zero, then that's the ideal answer and that is the ideal output that 40 00:03:18,980 --> 00:03:19,860 we are looking for. 41 00:03:20,450 --> 00:03:28,620 There are several ways that you can adjust your parameters to get the best result for the output. 42 00:03:29,090 --> 00:03:33,590 That's why we need to train our system in order to minimize the error. 43 00:03:33,890 --> 00:03:37,280 Let me just show it to you with the example. 44 00:03:37,670 --> 00:03:40,610 Suppose this is your house and this is your workplace. 45 00:03:42,140 --> 00:03:49,640 I can say there are infinite number of different ways to get from your home to your workplace and you 46 00:03:49,700 --> 00:03:52,800 choose one of them based on the situation. 47 00:03:52,820 --> 00:03:57,230 It's not usually the shortest, not always the shortest way. 48 00:03:57,500 --> 00:04:02,990 Sometimes if you go, lets me just call them road one to three. 49 00:04:03,770 --> 00:04:09,200 Sometimes if you choose route number three, then you can escape the traffic and you can get to your 50 00:04:09,200 --> 00:04:10,430 workplace faster. 51 00:04:11,240 --> 00:04:17,210 But there are infinite number of ways that you can get from your home to your workplace. 52 00:04:17,510 --> 00:04:21,790 Not all of them are actually a good way. 53 00:04:22,160 --> 00:04:28,820 But however, I can say there are infinite numbers of getting from your home to your workplace. 54 00:04:29,600 --> 00:04:30,930 Let's look at this one. 55 00:04:30,950 --> 00:04:32,510 This is another example. 56 00:04:32,660 --> 00:04:39,740 This is a very not very ideal, but this is a way for our system. 57 00:04:39,750 --> 00:04:40,660 It's also the same. 58 00:04:40,700 --> 00:04:45,920 There are infinite numbers of ways that we can adjust the number of layers we can add. 59 00:04:45,920 --> 00:04:53,410 Just a parameter is the way to biases, but only one of them is the best based number system. 60 00:04:53,930 --> 00:05:02,780 We need to train over a network and we are looking to find the best way to minimize this error and to 61 00:05:02,780 --> 00:05:07,550 have the closest result for why and why that.