1 00:00:00,420 --> 00:00:03,990 Hi and welcome to the chapter on fully connected layers. 2 00:00:04,380 --> 00:00:09,630 We'll explore what it is and what the purpose of a fully connected layer is in the CNN. 3 00:00:10,590 --> 00:00:12,690 So firstly, what is fully connected mean? 4 00:00:13,140 --> 00:00:17,940 Well, simply means all the nodes in one layer are connected to all the outputs in the next layer. 5 00:00:18,030 --> 00:00:20,550 I will describe that to you in the diagram shortly. 6 00:00:21,240 --> 00:00:26,730 However, just so you know it takes two to tree volume output from the previously and flattens it into 7 00:00:26,730 --> 00:00:30,630 a single vector and that's used for the inputs into the next layer. 8 00:00:31,110 --> 00:00:35,820 It's also sometimes called a dense layer, so this text is confusing. 9 00:00:36,030 --> 00:00:40,620 Let's take a look at the diagram and get a better understanding of what the fully connected layer does. 10 00:00:41,220 --> 00:00:44,120 So you can see these are feature maps here. 11 00:00:44,700 --> 00:00:45,720 Remember, we got features. 12 00:00:45,720 --> 00:00:51,600 We get feature maps from after applying the convolution filter to the image, and then we can apply 13 00:00:51,600 --> 00:00:53,550 some school to it and regional as well. 14 00:00:54,090 --> 00:00:55,380 So these are the outputs here. 15 00:00:56,310 --> 00:01:00,020 So what fully connected layer the EFSI layer does? 16 00:01:00,420 --> 00:01:07,530 It simply flattens all of these filters of all of these outputs into a single vector like this, so 17 00:01:07,530 --> 00:01:13,350 you can see one two three one six seven one eight seven hits in a row here and column here. 18 00:01:13,890 --> 00:01:16,290 And then the next filters added. 19 00:01:16,650 --> 00:01:19,800 And then the last filter is added again to the bottom. 20 00:01:20,010 --> 00:01:27,360 So you have this one dimensional vector with all the values from and from the feature maps before, 21 00:01:28,170 --> 00:01:31,290 and every note here is connected to the output nodes. 22 00:01:31,650 --> 00:01:36,690 These output nodes will deal with in the next section, but these output nodes are very important. 23 00:01:37,020 --> 00:01:43,680 And basically, we have all of these up these and this nodes here connected to each output. 24 00:01:43,740 --> 00:01:45,210 That's why there's so many lines here. 25 00:01:46,110 --> 00:01:49,050 So what's the purpose of this fully connected layer? 26 00:01:49,680 --> 00:01:56,460 Well, it compiles the data outputs extracted from the feature maps and previous layers to form the 27 00:01:56,460 --> 00:01:57,300 final output. 28 00:01:57,450 --> 00:01:58,770 But that doesn't tell you much. 29 00:01:58,780 --> 00:02:05,490 So effectively, imagine this it's an easy way of putting neural net to learn the nonlinear combinations 30 00:02:05,490 --> 00:02:06,370 of these features. 31 00:02:06,900 --> 00:02:09,330 Because remember, we have all of these features. 32 00:02:09,660 --> 00:02:10,990 This one could be a whisker. 33 00:02:11,010 --> 00:02:15,930 This one could be a teele, this one from the AI, and we have all of these values of indicates something 34 00:02:16,440 --> 00:02:17,180 we need to know. 35 00:02:17,190 --> 00:02:24,930 Find a way for the neural network to combine different groups of or different activations to and correlated 36 00:02:24,930 --> 00:02:26,100 to which output is which. 37 00:02:26,430 --> 00:02:28,080 That's what output nodes are for in the end. 38 00:02:28,470 --> 00:02:31,830 Don't worry, this will all fit together nicely in the end. 39 00:02:32,220 --> 00:02:35,460 I'll do a review of everything, and it will all make sense, I promise you. 40 00:02:36,270 --> 00:02:38,790 So that's it for the fully connected layer. 41 00:02:39,210 --> 00:02:44,550 Next, we take a look at the final list and then you're in the CNN, which is the soft max layer. 42 00:02:45,120 --> 00:02:46,710 So I'll see you in the next section. 43 00:02:46,890 --> 00:02:47,340 Thank you.