1
00:00:07,750 --> 00:00:13,300
In this video, we’ll introduce you to machine
learning and deep learning models.

2
00:00:13,300 --> 00:00:18,700
Data contains a wealth of information that
can be used to solve certain types of problems.

3
00:00:18,700 --> 00:00:24,420
Traditional data analysis approaches, such
as a person manually inspecting the data or

4
00:00:24,420 --> 00:00:29,880
a specialized computer program that automates
the human analysis, quickly reach their limits

5
00:00:29,880 --> 00:00:35,570
due to the amount of data to be analyzed or
the complexity of the problem.

6
00:00:35,570 --> 00:00:40,520
Machine learning uses algorithms – also
known as ”models” - to identify patterns

7
00:00:40,520 --> 00:00:41,730
in the data.

8
00:00:41,730 --> 00:00:47,960
The process by which the model learns these
patterns from data is called “model training."

9
00:00:47,960 --> 00:00:52,010
Once a model is trained, it can then be used
to make predictions.

10
00:00:52,010 --> 00:00:57,129
When the model is presented with new data,
it tries to make predictions or decisions

11
00:00:57,129 --> 00:01:01,650
based on the patterns it has learned from
past data.

12
00:01:01,650 --> 00:01:06,850
Machine learning models can be divided into
three basic classes: supervised learning,

13
00:01:06,850 --> 00:01:11,460
unsupervised learning, and reinforcement learning.

14
00:01:11,460 --> 00:01:16,130
Supervised learning is one of the most commonly
used type of machine learning models.

15
00:01:16,130 --> 00:01:21,930
In supervised learning, a human provides input
data and the correct outputs.

16
00:01:21,930 --> 00:01:27,689
The model tries to identify relationships
and dependencies between the input data and

17
00:01:27,689 --> 00:01:29,930
the correct output.

18
00:01:29,930 --> 00:01:35,140
Generally speaking, supervised learning is
used to solve regression and classification

19
00:01:35,140 --> 00:01:36,140
problems.

20
00:01:36,140 --> 00:01:40,180
Let’s look at an example for each problem
type:

21
00:01:40,180 --> 00:01:44,119
Regression models are used to predict a numeric,
or “real," value.

22
00:01:44,119 --> 00:01:50,689
For example, given information about past
home sales, such as geographic location, size,

23
00:01:50,689 --> 00:01:56,350
number of bedrooms, and sales price, you can
train a model to predict the estimated sales

24
00:01:56,350 --> 00:02:02,720
price for other homes with similar characteristics.

25
00:02:02,720 --> 00:02:06,520
Classification models are used to predict
whether something belongs to a category, or

26
00:02:06,520 --> 00:02:07,520
“class."

27
00:02:07,520 --> 00:02:13,790
For example, given a set of emails along with
a designation of whether or not they are considered

28
00:02:13,790 --> 00:02:19,110
spam, an algorithm can be trained to identify
unsolicited emails.

29
00:02:19,110 --> 00:02:23,510
In unsupervised learning, the data is not
labelled by a human.

30
00:02:23,510 --> 00:02:28,620
The models must analyze the data and try to
identify patterns and structure within the

31
00:02:28,620 --> 00:02:34,610
data based only on the characteristics of
the data itself.

32
00:02:34,610 --> 00:02:39,650
Clustering and anomaly detection are two examples
of this learning style.

33
00:02:39,650 --> 00:02:44,610
Clustering models are used to divide each
record of a data set into one of a small number

34
00:02:44,610 --> 00:02:46,580
of similar groups.

35
00:02:46,580 --> 00:02:50,730
An example of a clustering model could be
providing purchase recommendations for an

36
00:02:50,730 --> 00:02:57,959
e-commerce store based on past shopping behavior
and the content of a shopping basket.

37
00:02:57,959 --> 00:03:04,220
Anomaly detection identifies outliers in a
data set, such as fraudulent credit card transactions

38
00:03:04,220 --> 00:03:07,500
or suspicious online log-in attempts.

39
00:03:07,500 --> 00:03:12,349
The third type of learning, reinforcement
learning, is loosely based on the way human

40
00:03:12,349 --> 00:03:15,440
beings and other organisms learn.

41
00:03:15,440 --> 00:03:17,420
Think about a mouse in a maze.

42
00:03:17,420 --> 00:03:21,400
If the mouse gets to the end of the maze it
gets a piece of cheese.

43
00:03:21,400 --> 00:03:24,250
This is the “reward” for completing a
task.

44
00:03:24,250 --> 00:03:28,310
The mouse learns – through trial and error
– how to get through the maze to get as

45
00:03:28,310 --> 00:03:31,230
much cheese as it can.

46
00:03:31,230 --> 00:03:37,010
In a similar way, a reinforcement learning
model learns the best set of actions to take,

47
00:03:37,010 --> 00:03:42,580
given its current environment, in order to
get the most reward over time.

48
00:03:42,580 --> 00:03:47,170
This type of learning has recently been very
successful in beating the best human players

49
00:03:47,170 --> 00:03:53,489
in games such as go, chess, and popular strategy
video games.

50
00:03:53,489 --> 00:03:56,849
Deep learning is a specialized type of machine
learning.

51
00:03:56,849 --> 00:04:02,239
It refers to a general set of models and techniques
that tries to loosely emulate the way the

52
00:04:02,239 --> 00:04:05,700
human brain solves a wide range of problems.

53
00:04:05,700 --> 00:04:11,930
It is commonly used to analyze natural language,
both spoken and text, as well as images, audio,

54
00:04:11,930 --> 00:04:16,860
and video, to forecast time series data and
much more.

55
00:04:16,860 --> 00:04:21,580
Deep learning has had a lot of recent success
in these and other areas and is therefore

56
00:04:21,580 --> 00:04:27,300
becoming an increasingly popular and important
tool for data science.

57
00:04:27,300 --> 00:04:32,940
Deep learning typically requires very large
data sets of labeled data to train a model,

58
00:04:32,940 --> 00:04:38,370
is compute-intensive, and usually requires
special purpose hardware to achieve acceptable

59
00:04:38,370 --> 00:04:40,070
training times.

60
00:04:40,070 --> 00:04:44,800
You can build a custom deep learning model
from scratch or use pre-trained models from

61
00:04:44,800 --> 00:04:48,720
public model repositories.

62
00:04:48,720 --> 00:04:54,280
Deep learning models are implemented using
popular frameworks such as TensorFlow, PyTorch,

63
00:04:54,280 --> 00:04:55,780
and Keras.

64
00:04:55,780 --> 00:05:01,210
Deep learning frameworks typically provide
a Python API, and many support other programming

65
00:05:01,210 --> 00:05:05,330
languages, such as C++ and JavaScript.

66
00:05:05,330 --> 00:05:10,009
You can download pre-trained state-of-the-art
models from repositories that are commonly

67
00:05:10,009 --> 00:05:13,270
referred to as model "zoos."

68
00:05:13,270 --> 00:05:20,449
Popular model zoos include those provided
by TensorFlow, PyTorch, Keras, and ONNX.

69
00:05:20,449 --> 00:05:25,090
Models are also published by academic and
commercial research groups.

70
00:05:25,090 --> 00:05:30,440
While it is beyond the scope of this video
to explain in detail how you would go about

71
00:05:30,440 --> 00:05:36,090
building a model, let’s briefly outline
the high-level tasks using an example.

72
00:05:36,090 --> 00:05:41,699
Assume you want to enable an application to
identify objects in images by training a deep

73
00:05:41,699 --> 00:05:43,229
learning model.

74
00:05:43,229 --> 00:05:48,270
First, you collect and prepare data that will
be used to train a model.

75
00:05:48,270 --> 00:05:52,620
Data preparation can be a time-consuming and
labor-intensive process.

76
00:05:52,620 --> 00:05:58,000
In order to train a model to detect objects
in images, you need to label the raw training

77
00:05:58,000 --> 00:06:04,410
data by, for example, drawing bounding boxes
around objects and labeling them.

78
00:06:04,410 --> 00:06:10,290
Next, you build a model from scratch or select
an existing model that might be well suited

79
00:06:10,290 --> 00:06:14,340
for the task from a public or private resource.

80
00:06:14,340 --> 00:06:17,280
You then train the model on your prepared
data.

81
00:06:17,280 --> 00:06:22,259
During training, your model learns from the
labeled data how to identify objects that

82
00:06:22,259 --> 00:06:24,470
are depicted in an image.

83
00:06:24,470 --> 00:06:29,580
Once training has commenced, you analyze the
training results and repeat the process until

84
00:06:29,580 --> 00:06:33,069
the trained model performance meets your requirements.

85
00:06:33,069 --> 00:06:37,830
When the trained model performs as desired,
you deploy it to make it available to your

86
00:06:37,830 --> 00:06:41,069
applications.

87
00:06:41,069 --> 00:06:45,490
In this video, you’ve learned about machine
learning and deep learning, what they are

88
00:06:45,490 --> 00:06:49,449
used for, and where to find open source models.

89
00:06:49,449 --> 00:06:54,740
In the next video, we’ll introduce you to
the Model Asset eXchange, a curated collection

90
00:06:54,740 --> 00:06:58,580
of ready-to-use and customizable deep learning
models.