1
00:00:00,480 --> 00:00:00,800
Hello.

2
00:00:00,810 --> 00:00:03,070
Welcome back.

3
00:00:03,090 --> 00:00:10,200
Early on in the course we saw how to update the weight parameters using a simple method.

4
00:00:10,230 --> 00:00:17,000
Our error or lost function in that lesson was the predicted value minus the expected value squared.

5
00:00:17,010 --> 00:00:24,600
We introduced Delta which we called the Pure aero and we expressed it by simply predicted value minus

6
00:00:24,660 --> 00:00:25,740
expected value.

7
00:00:25,750 --> 00:00:34,160
Without this word we went on to compute the weight Delta by multiplying the delta by the input.

8
00:00:34,170 --> 00:00:40,470
And we said in order to prevent over scaling we had to introduce a new constant known as the learning

9
00:00:40,470 --> 00:00:43,170
rate or Alpha.

10
00:00:43,170 --> 00:00:50,190
We then updated the weight by multiplying the weight Delta by Alpha and then subtracting the answer

11
00:00:50,190 --> 00:00:54,960
from the current weight value as we can see over here.

12
00:00:55,050 --> 00:01:02,280
This is what we showed before to demonstrate the concept of learning and updating the weight based on

13
00:01:02,280 --> 00:01:03,420
what is learned.

14
00:01:04,590 --> 00:01:10,410
Let's see how that same concept can be applied here to our cuts or no cut.

15
00:01:10,410 --> 00:01:17,400
Simple logistic regression example our cost function is expressed by this equation as we saw some lessons

16
00:01:17,400 --> 00:01:17,950
ago.

17
00:01:18,150 --> 00:01:25,880
Our network has four parameters to be one and B one are the weight and biases of Layer 1.

18
00:01:25,920 --> 00:01:33,330
What these mean is these are the weight and biases which are multiplied and added to the input to derive

19
00:01:33,360 --> 00:01:42,720
the result Z subscript 1 0 subscript 2 Zi subscript 3 and c subscript 4 which are ultimately passed

20
00:01:42,720 --> 00:01:50,610
through the activation function to obtain a subscript 1 a subscript to a subscript 3 and a subscript

21
00:01:50,610 --> 00:02:00,130
4 w superscript 1 and b superscript 1 are therefore known as the parameters of Layer 1 because they

22
00:02:00,130 --> 00:02:02,470
give us the result of Layer 1.

23
00:02:02,470 --> 00:02:09,370
We also have w superscript 2 and b superscript 2 these parameters take the result hopefully a 1 and

24
00:02:09,370 --> 00:02:15,370
transform it to give us the result of Layer 2 using the same method I just described.

25
00:02:15,370 --> 00:02:21,550
So you know what's normally a new route network we have 4 parameters 2 for Layer 1 and 2 for layer to

26
00:02:21,880 --> 00:02:27,460
remember as I mentioned earlier when counting the layers of a neural network we don't count the inputs

27
00:02:27,460 --> 00:02:29,280
layer right.

28
00:02:29,290 --> 00:02:35,530
So back to the cost function we have to find the cost with respect to these four parameters so that

29
00:02:35,530 --> 00:02:42,040
we can reduce the error between what they currently are and what they need to be in order to give us

30
00:02:42,430 --> 00:02:49,840
a predicted value which is as close to the expected value as possible we will update w 1 by computing

31
00:02:50,230 --> 00:02:58,690
B.J. over D.W. 1 and then multiplying the result by our lending rate alpha and then subtracting the

32
00:02:58,690 --> 00:03:09,120
results from the current w 1 value to get the new w 1 we will update B1 by computing DG divided by DP

33
00:03:09,130 --> 00:03:15,940
1 then multiplying the resource by our Len in great Alpha and then subtracting the results from the

34
00:03:15,940 --> 00:03:23,500
current we want to get the new B1 like we can see right here we do the same thing to update W2 and b

35
00:03:23,530 --> 00:03:27,310
to this is how we shall update our parameters.

36
00:03:27,520 --> 00:03:32,620
So that's all there is for this lesson if you have any questions at all just let me know and I'll see

37
00:03:32,620 --> 00:03:33,550
you in the next lesson.