Would you like to inspect the original subtitles? These are the user uploaded subtitles that are being translated:
1
00:00:00,390 --> 00:00:04,590
After we have a detailed overview of what happens inside convolutional neural networks.
2
00:00:04,590 --> 00:00:09,750
Now, we will have an insight for simple case illustration of how classification works using CNN.
3
00:00:10,770 --> 00:00:16,830
This is the overview of the case and input images processed by CNN architecture and classified as a
4
00:00:16,830 --> 00:00:17,850
cross or a circle.
5
00:00:18,720 --> 00:00:24,210
Now we will go trough on how exactly a classification happens on this simple convolutional architecture.
6
00:00:25,290 --> 00:00:30,900
The input image is a simple black and white image, whereas white pixels represented by pixel one and
7
00:00:30,900 --> 00:00:33,270
black pixel represented by pixel zero.
8
00:00:34,370 --> 00:00:39,560
Following the conversion was carried out using a number of filters in the convolution layers, producing
9
00:00:39,560 --> 00:00:40,760
some filter maps.
10
00:00:41,770 --> 00:00:45,340
After completing the convolution, the activation function is used.
11
00:00:46,150 --> 00:00:49,860
We to maintain the positive value while changing the negative value to zero.
12
00:00:49,880 --> 00:00:51,470
As was previously explained.
13
00:00:52,610 --> 00:00:55,010
The fetal map then comes into the pooling layer.
14
00:00:55,980 --> 00:01:01,230
In this example, we use max palling with radicals, one from the pooling process.
15
00:01:01,230 --> 00:01:03,300
We will have done simple feature maps.
16
00:01:03,960 --> 00:01:08,430
The feature maps is then transformed from a multi-dimensional array to a flat vector so that it can
17
00:01:08,430 --> 00:01:10,650
be used as input for the fully connected layer.
18
00:01:11,430 --> 00:01:15,360
In this example, the flattened result is connected to two output neurons.
19
00:01:16,080 --> 00:01:18,570
Each connection on a neuron has an associated way.
20
00:01:19,500 --> 00:01:23,520
As an additional note, the weights value is obtained during the training process.
21
00:01:24,560 --> 00:01:29,960
The soft makes activation function are applied to the two neurons in order to determine the class probability.
22
00:01:30,650 --> 00:01:35,300
CNN predicts that the input image has a cross shape because the probability of a cross class is the
23
00:01:35,300 --> 00:01:37,760
highest based on the image that was actually seen.
24
00:01:38,630 --> 00:01:43,220
After understanding how deep learning works, now is the time for you to understand how all of these
25
00:01:43,220 --> 00:01:44,180
seven works.
26
00:01:44,840 --> 00:01:45,590
See you then.
2736
Can't find what you're looking for?
Get subtitles in any language from opensubtitles.com, and translate them here.