Introduction to Machine Learning — Transcript

Full transcript

1. 0:00[Music]
2. 0:12we are currently living in a data era
3. 0:15where a vast amount of data is collected
4. 0:17and stored every day in the face of this
5. 0:21growing quantity of data machine
6. 0:23learning methods have become inescapable
7. 0:26so much so that you probably use them
8. 0:29dozens of times a day without even
9. 0:31noticing let's start with an example of
10. 0:35an everyday machine learning
11. 0:36contribution for millions of users the
12. 0:39algorithm behind Facebook's newsfeed
13. 0:42Facebook uses machine learning to
14. 0:44exploit users data and feedback to
15. 0:47personalize their feeds if you like a
16. 0:50post or stop scrolling to read something
17. 0:52the algorithm learns from this and
18. 0:54starts to populate your feed with
19. 0:56further similar content this learning is
20. 0:59done continuously and so the material
21. 1:02suggested in your newsfeed evolves with
22. 1:04your preferences making your user
23. 1:06experience more enjoyable this is only
24. 1:10one example there are many others Apple
25. 1:14can recognize your friends face in the
26. 1:16photo you just took Amazon echo
27. 1:18understands you and can answer your
28. 1:20questions your vacuum can even navigate
29. 1:23its way around your house while Netflix
30. 1:26is recommending videos that match your
31. 1:28profile machine learning has become a
32. 1:31massive part of our daily lives and it's
33. 1:33not going anywhere soon but what is
34. 1:36machine learning exactly what's behind
35. 1:39these magical looking algorithms and how
36. 1:42do they use data to work so well
37. 1:46formerly machine learning is the science
38. 1:49of getting computers to realize a task
39. 1:51without being explicitly programmed in
40. 1:54other words the big difference between
41. 1:57classical and machine learning
42. 1:58algorithms lies in the way we define
43. 2:01them
44. 2:03classical algorithms are given exact and
45. 2:06complete rules to complete a task
46. 2:10machine learning algorithms are given
47. 2:12general guidelines that define the model
48. 2:15along with data this data should contain
49. 2:18the missing information necessary for
50. 2:20the model to complete the task so a
51. 2:23machine learning algorithm can
52. 2:24accomplish its task when the model has
53. 2:27been adjusted with respect to the data
54. 2:29we say that we fit the model on the data
55. 2:33or that the model has to be trained on
56. 2:35the data let's illustrate this with a
57. 2:39simple example let's say we want to
58. 2:43predict the price of a house based on
59. 2:45the size of the house the size of its
60. 2:48garden and the number of rooms it has we
61. 2:55could try to build a classical algorithm
62. 2:57that answers this problem this algorithm
63. 3:00would have to take the three house
64. 3:02features and return the predicted price
65. 3:04based on an explicit rule in this
66. 3:09example the exact house pricing formula
67. 3:11has to be known and coded explicitly but
68. 3:14in practice this formula is often not
69. 3:17known on the other hand we could build a
70. 3:24machine learning algorithm first such
71. 3:27algorithms would define a model that can
72. 3:29be an incomplete formula created from
73. 3:32our limited knowledge then the model
74. 3:35would be adjusted by training on given
75. 3:38housing prices examples doing so we
76. 3:43combine a model with some data
77. 3:48in general machine learning is
78. 3:51incredibly useful for difficult tasks
79. 3:54when we have incomplete information or
80. 3:56information that's too complex to be
81. 3:58coded by hand in these cases we can give
82. 4:02the information we have available to our
83. 4:04model and let this one learn the missing
84. 4:07information that it needs by itself the
85. 4:10algorithm will then use statistical
86. 4:12techniques to extract the missing
87. 4:14knowledge directly from the data the two
88. 4:20main categories of machine learning
89. 4:21techniques are supervised learning and
90. 4:24unsupervised learning in supervised
91. 4:28learning we want to get a model to
92. 4:30predict the label of data based on their
93. 4:33features in order to learn mapping
94. 4:37between features and labels the model
95. 4:39has to be fitted on given examples of
96. 4:42features with they're related labels we
97. 4:45say that the model is trained on the
98. 4:48labelled data set
99. 4:51predicted labels can be numbers or
100. 4:53categories for example we could be
101. 4:56building a model that predicts the price
102. 4:58of a house implying we would want to
103. 5:01predict a label that's a number in this
104. 5:04case we would talk about a regression
105. 5:06model otherwise we might also want to
106. 5:10define a model that predicts a category
107. 5:12like cat or not cat based on given
108. 5:16features in this situation we would talk
109. 5:18about a classification model in
110. 5:23unsupervised learning we want to define
111. 5:25a model that reveals structures in some
112. 5:28data that are described only by their
113. 5:30features but with no labels for example
114. 5:34unsupervised learning algorithms can
115. 5:37help answer questions like are there
116. 5:39groups among my data or is there any way
117. 5:42to simplify the description of my data
118. 5:44the model can look for different kinds
119. 5:49of underlying structures in the data if
120. 5:52it tries to find groups among the data
121. 5:54we would talk about a clustering model
122. 5:57an example of a clustering model would
123. 6:01be a model that segments customers of a
124. 6:04company based on their profiles
125. 6:07otherwise if we have a model that
126. 6:09transforms data and represents them with
127. 6:12a smaller number of features we would
128. 6:14talk about a dimension reduction model
129. 6:16an example of this would be a model that
130. 6:19summarizes the multiple technical
131. 6:21characteristics of some cars into a few
132. 6:24main indicators in summary supervised
133. 6:32learning models associate a label with
134. 6:35each data point described by its
135. 6:36features whereas unsupervised learning
136. 6:39models find structures among all the
137. 6:41data points
138. 6:43in a sense supervised learning is
139. 6:46similar to learning the names of fruits
140. 6:48from a picture book you associate the
141. 6:52characteristics of the fruit the
142. 6:54features with the names written on the
143. 6:56page the label classical examples of
144. 7:00supervised learning algorithms are
145. 7:02linear regression logistic regression
146. 7:04support vector machines neural networks
147. 7:07and so on unsupervised learning on the
148. 7:13other hand is like taking the same fruit
149. 7:15picture book analyzing all of the fruits
150. 7:18to detect patterns and then deciding to
151. 7:20group fruits by color and size classical
152. 7:25examples of unsupervised learning
153. 7:27algorithms are k-means clustering
154. 7:30hierarchical clustering principal
155. 7:32component analysis autoencoders and so
156. 7:35on let's conclude this video by
157. 7:41mentioning that machine learning is not
158. 7:43that new and that many of the algorithms
159. 7:45driving today's applications have been
160. 7:48around for years
161. 7:50nevertheless some major advances took
162. 7:53place along the way we have built
163. 7:55datasets larger than ever before we have
164. 7:59increased our computation power and we
165. 8:02have imagined new cutting-edge models if
166. 8:06these advances have already made it
167. 8:08possible to approach and in some cases
168. 8:10even exceed human abilities across many
169. 8:13tasks there is also no doubt that we are
170. 8:15only scratching the surface of what's
171. 8:18possible
172. 8:25we really hope you enjoyed this video do
173. 8:28not hesitate to leave feedback and tell
174. 8:30us in the comments section if there are
175. 8:32any topics you would be interested in
176. 8:34for coming videos thanks for watching
177. 8:38and see you soon on towards data science