1
00:00:00,000 --> 00:00:03,658
Hi and welcome back to our last section

2
00:00:03,659 --> 00:00:07,348
in this training. I would like to recap the things

3
00:00:07,349 --> 00:00:10,800
we covered so far for creating an end-to-end story.

4
00:00:11,570 --> 00:00:16,442
Artificial intelligence is an umbrella term for the fundamental

5
00:00:16,443 --> 00:00:20,714
idea that some complex intellectual tasks that are performed

6
00:00:20,715 --> 00:00:23,570
by humans can also be performed by machines.

7
00:00:23,571 --> 00:00:28,802
When a machine can mimic or complex cognitive functions

8
00:00:28,803 --> 00:00:32,540
like identifying an object in a picture or

9
00:00:32,541 --> 00:00:35,986
recognize a human voice and many other complex

10
00:00:35,987 --> 00:00:41,310
tasks, it is described as artificial intelligence.

11
00:00:41,311 --> 00:00:44,960
Still machines can perform complex tasks using

12
00:00:44,961 --> 00:00:48,832
pre-programmed rules, but in many cases this

13
00:00:48,833 --> 00:00:52,132
traditional programming approach cannot scale up to

14
00:00:52,133 --> 00:00:55,490
handle very complex tasks.

15
00:00:55,491 --> 00:00:57,092
Something is missing here.

16
00:00:57,093 --> 00:00:59,994
The missing part of AI is the flexibility

17
00:00:59,995 --> 00:01:02,888
to learn, which is the next phase of

18
00:01:02,889 --> 00:01:06,638
the AI evolution path moving to the concept

19
00:01:06,639 --> 00:01:10,184
of machine learning, machines that can learn.

20
00:01:10,185 --> 00:01:13,528
So instead of using pre-programmed rules, we

21
00:01:13,529 --> 00:01:16,200
let the machines learn from the data,

22
00:01:16,730 --> 00:01:20,172
extract knowledge automatically and then use the

23
00:01:20,173 --> 00:01:23,110
knowledge to perform a complex task.

24
00:01:23,770 --> 00:01:26,540
The knowledge in a machine learning system

25
00:01:26,541 --> 00:01:31,062
is basically some mathematical transformation mapping functions

26
00:01:31,063 --> 00:01:34,326
that are encapsulated in layers.

27
00:01:34,327 --> 00:01:37,302
We talked about shallow learning, meaning

28
00:01:37,303 --> 00:01:39,856
using very small amount of layers. Okay.

29
00:01:39,857 --> 00:01:42,058
This is the knowledge, the brain in the machine

30
00:01:42,059 --> 00:01:45,946
learning system or using deep learning with many layers.

31
00:01:45,947 --> 00:01:49,466
The parameters of those mapping functions

32
00:01:49,467 --> 00:01:53,144
are identified automatically during the training

33
00:01:53,145 --> 00:01:57,490
process by a specific optimization algorithm.

34
00:01:58,150 --> 00:02:00,344
We also talked about the options

35
00:02:00,345 --> 00:02:02,904
to train a machine learning system.

36
00:02:02,905 --> 00:02:04,680
The first very common option

37
00:02:04,681 --> 00:02:06,780
is called supervised learning.

38
00:02:06,781 --> 00:02:09,788
In this method, we need to supervise the

39
00:02:09,789 --> 00:02:14,354
learning process by providing the machine learning algorithm

40
00:02:14,355 --> 00:02:18,012
labeled examples as a training dataset.

41
00:02:18,013 --> 00:02:20,848
This training dataset will be used to train a

42
00:02:20,849 --> 00:02:24,928
model that can be used by an ML system in

43
00:02:24,929 --> 00:02:30,900
a production environment, something that is called inference.Training

44
00:02:30,901 --> 00:02:34,532
a model is a complex process of selecting the

45
00:02:34,533 --> 00:02:41,412
relevant algorithm, tuning the input features, cleaning the data,

46
00:02:41,413 --> 00:02:45,562
testing the performance of the model, trying to overcome

47
00:02:45,563 --> 00:02:49,110
two main challenges underfitting and overfitting,

48
00:02:49,111 --> 00:02:52,536
and the final goal of a trained model is to

49
00:02:52,537 --> 00:02:57,960
be well generalized and to handle new unseen data.

50
00:02:58,810 --> 00:03:01,708
There are two typical tasks that can be

51
00:03:01,709 --> 00:03:07,330
handled under supervised learning classification and regression.

52
00:03:07,331 --> 00:03:10,982
Classification is a very popular use case in supervised

53
00:03:10,983 --> 00:03:14,166
learning and it is used to classify the input

54
00:03:14,167 --> 00:03:19,670
stream like image classifier or spam classifier.

55
00:03:19,671 --> 00:03:22,522
We can build a classifier using shallow

56
00:03:22,523 --> 00:03:25,690
learning or using deep learning algorithm.

57
00:03:25,691 --> 00:03:27,908
The second very common method in

58
00:03:27,909 --> 00:03:30,650
supervised learning is called regression.

59
00:03:30,651 --> 00:03:33,806
It is a very straightforward method to predict

60
00:03:33,807 --> 00:03:37,864
a continuous number based on historical data.

61
00:03:37,865 --> 00:03:41,976
The concept of regression analysis is widely used

62
00:03:41,977 --> 00:03:45,996
for data analysis, but in addition it is

63
00:03:45,997 --> 00:03:49,276
also one of the most basic form of

64
00:03:49,277 --> 00:03:54,892
machine learning. Moving next to unsupervised learning.

65
00:03:54,893 --> 00:03:58,166
One of the main challenges when using supervised

66
00:03:58,167 --> 00:04:01,216
learning is to get labeled dataset because

67
00:04:01,217 --> 00:04:04,304
in most practical application we don't have it,

68
00:04:04,305 --> 00:04:07,790
we will have an unlabeled dataset.

69
00:04:07,791 --> 00:04:10,548
Unsupervised learning is learning without a

70
00:04:10,549 --> 00:04:13,492
teacher supervising the learning process.

71
00:04:13,493 --> 00:04:17,034
The goal is to identify automatically

72
00:04:17,035 --> 00:04:21,440
meaningful patterns in unlabeled data.

73
00:04:22,690 --> 00:04:24,756
It is used for two main

74
00:04:24,757 --> 00:04:29,990
fundamental task clustering and dimension reduction.

75
00:04:29,991 --> 00:04:32,750
Clustering is about summarizing and grouping

76
00:04:32,751 --> 00:04:36,888
similar instances together into clusters okay.

77
00:04:36,889 --> 00:04:39,522
Like cluster number 1, 2, 3, etc.

78
00:04:39,523 --> 00:04:42,220
It is widely used by search

79
00:04:42,221 --> 00:04:48,010
engines, customer segmentation, application image segmentation,

80
00:04:48,011 --> 00:04:50,390
and simple data analysis.

81
00:04:51,050 --> 00:04:54,482
Dimension reduction is about reducing the complexity

82
00:04:54,483 --> 00:04:57,612
of the input data while maintaining the

83
00:04:57,613 --> 00:04:59,996
structure and usefulness of the data.

84
00:04:59,997 --> 00:05:03,300
This method is sometimes used as a

85
00:05:03,301 --> 00:05:08,698
preprocess stage to the data before feeding

86
00:05:08,699 --> 00:05:11,594
it into a supervised learning algorithm.

87
00:05:11,595 --> 00:05:15,384
The third learning option is reinforcement learning.

88
00:05:15,385 --> 00:05:18,232
It is used in applications that the

89
00:05:18,233 --> 00:05:21,582
machine must make a sequence of decisions

90
00:05:21,583 --> 00:05:25,858
while interacting with the outside environment.

91
00:05:25,859 --> 00:05:29,170
Those decisions are coming with positive or negative

92
00:05:29,171 --> 00:05:34,170
consequences, which will be translated by the agent

93
00:05:34,171 --> 00:05:36,876
as a feedback loop to learn what is

94
00:05:36,877 --> 00:05:38,944
working and what is not working.

95
00:05:38,945 --> 00:05:43,830
The cumulative knowledge that the agent is creating

96
00:05:43,831 --> 00:05:46,560
on how to achieve a specific goal is

97
00:05:46,561 --> 00:05:51,092
reinforced again and again based on experience, okay,

98
00:05:51,093 --> 00:05:54,240
the same way as humans are learning from experience.

99
00:05:55,170 --> 00:05:57,380
So those are the three

100
00:05:57,381 --> 00:06:00,100
options of machine learning system.

101
00:06:00,101 --> 00:06:04,356
It was a quick recap to connect the dots. That's it.

102
00:06:04,357 --> 00:06:07,172
I want to thank you for watching this training.

103
00:06:07,173 --> 00:06:09,172
I hope that you enjoyed it and

104
00:06:09,173 --> 00:06:11,636
learned some interesting things along the way.

105
00:06:11,637 --> 00:06:15,044
It will be awesome and useful if you can

106
00:06:15,045 --> 00:06:18,268
can rate the course and share your experience.

107
00:06:18,269 --> 00:06:20,994
I'm planning to create multiple courses

108
00:06:20,995 --> 00:06:22,972
under the concept of machine learning

109
00:06:22,973 --> 00:06:25,618
and also on other interesting topics.

110
00:06:25,619 --> 00:06:28,188
So check out the bonus lecture at the end of

111
00:06:28,189 --> 00:06:32,588
the training and see which courses are already available today.

112
00:06:32,589 --> 00:06:36,730
I hope to see you again in my next training courses.

113
00:06:36,731 --> 00:06:37,910
Bye bye.

114
00:06:37,911 --> 00:06:40,490
[No audio]