WEBVTT
Kind: captions
Language: en

00:00:14.540 --> 00:00:15.740
Hello everybody.

00:00:15.780 --> 00:00:22.940
Today I want to introduce Artificial Neural Networks and their applications.

00:00:23.660 --> 00:00:28.280
This is an outline of
today's course including Introduction,

00:00:28.840 --> 00:00:33.100
Fundamental Concept of Artificial Neural Networks,

00:00:33.700 --> 00:00:35.940
Brief Introduction of Deep Learning,

00:00:36.480 --> 00:00:37.800
Applications.

00:00:39.140 --> 00:00:42.440
Now I'm going to start with introduction.

00:00:43.300 --> 00:00:49.260
Please remind that what do Fuzzy sets and fuzzy logic do?

00:00:50.240 --> 00:01:01.220
The answer is Fuzzy sets and fuzzy logic imitate the way the brain deals with inexact information.

00:01:02.040 --> 00:01:06.280
In contrast, artificial neural networks

00:01:06.520 --> 00:01:14.800
are modeled to imitate the physical architecture of the brain.

00:01:16.460 --> 00:01:27.620
Artificial neural networks have the ability to classify, store, recall and associate information or patterns.

00:01:28.420 --> 00:01:35.660
I will say that artificial neural networks have the learning ability.

00:01:36.560 --> 00:01:44.860
Now I am going to introduce the Fundamental Concept of Artificial Neural Networks.

00:01:45.260 --> 00:01:53.660
Before mentioning about the artificial neural networks, we have to see what is neuron.

00:01:54.400 --> 00:01:58.460
The figure on the left is the real neuron.

00:01:59.160 --> 00:02:08.580
The dendrite can receive the electrochemical stimulation from other neural cells.

00:02:09.160 --> 00:02:21.960
The cell body can do some process for the stimulation and transmit it to different neurons through Axon.

00:02:23.040 --> 00:02:27.500
The figure on the right is the artificial neuron.

00:02:28.680 --> 00:02:35.860
Here the x is the input and the w is the weight for the input.

00:02:36.360 --> 00:02:44.240
The weighted inputs are processed by the processing element to get the output y.

00:02:44.860 --> 00:02:47.980
You can see that the artificial neuron

00:02:48.880 --> 00:02:51.880
works just like the real one.

00:02:52.780 --> 00:02:58.240
By different kinds of connection of artificial neurons,

00:02:58.500 --> 00:03:03.800
we can obtain different kinds of artificial neural networks.

00:03:04.620 --> 00:03:08.680
Every neural networks has its own learning rule.

00:03:09.300 --> 00:03:15.180
The learning rules can be categorized into three major types,

00:03:15.780 --> 00:03:17.040
Supervised Learning,

00:03:17.040 --> 00:03:18.620
Unsupervised Learning

00:03:18.940 --> 00:03:21.260
and Reinforcement Learning

00:03:22.040 --> 00:03:27.740
For supervised learning, we have to teach the artificial neural network

00:03:28.220 --> 00:03:30.760
what is the correct answer.

00:03:31.840 --> 00:03:39.080
Unsupervised learning is like that children will cluster similar things

00:03:39.540 --> 00:03:41.480
without our teaching.

00:03:42.620 --> 00:03:48.340
Reinforcement learning is to teach the artificial neural network

00:03:49.060 --> 00:03:53.600
for taking action to get maximum reward.

00:03:54.280 --> 00:04:02.260
Now Let’s  go a little deeper in artificial neural networks that is deep learning.

00:04:02.600 --> 00:04:07.520
We can see here Deep learning is a subset of AI.

00:04:08.000 --> 00:04:12.620
Deep learning is implemented by deep neural networks

00:04:13.100 --> 00:04:17.840
which are more complicated multi-layer neural networks.

00:04:18.520 --> 00:04:26.560
Recently, the dramatic increase in computer power makes the deep learning possible.

00:04:27.120 --> 00:04:34.280
We can say that the deep learning has brought us to a new generation of AI.

00:04:34.760 --> 00:04:42.720
I think the ALPHOGO is one of the most representative applications of deep learning.

00:04:43.660 --> 00:04:46.300
It was considered to be

00:04:46.300 --> 00:04:50.080
almost impossible that a computer program

00:04:50.400 --> 00:04:56.000
could beat the GO world champion before the birth of ALPHOGO.

00:04:56.680 --> 00:05:01.520
We can see here the ALPHOGO applies a deep neural network.

00:05:02.420 --> 00:05:08.240
Finally we can see some applications of Artificial Neural Networks

00:05:08.940 --> 00:05:16.380
Artificial neural net works can be applied for AI to play video games.

00:05:17.060 --> 00:05:21.600
It can also be use for Object Detection.

00:05:21.820 --> 00:05:28.980
The artificial neural networks YOLO can detect objects in real time.

00:05:29.740 --> 00:05:37.040
It’s a difficult task for robot to fetch objects of random shapes.

00:05:37.300 --> 00:05:43.000
Artificial neural networks can be applied to detect

00:05:43.520 --> 00:05:46.400
Robotic Grasps for fetching them.

00:05:47.340 --> 00:05:57.480
Moreover, Artificial neural net works can be applied for decision making, recognition, etc.