Applied Edge AI: Deep Learning Outside of the Cloud

1. 00:00Hello. A Tractica report predicts that edgeAI device
2. 00:05shipments will increase from 161 million units in 2018 to 2.6 billion units worldwide in 2025.
3. 00:15You may be curious about which application areas edgeAI
4. 00:19will have a substantial impact on. In this video
5. 00:23we will do a brief analysis.
6. 00:275G and edge computing are two inextricably linked technology.
7. 00:35They both significantly improve the performance of applications and enable huge amount of data to be processed in real time.
8. 00:445G increases speed by up to 20 times that of 4G.
9. 00:49Whereas age computing reduced latency by bringing computing capability into the network closer to the end user and there
10. 00:59are still major unknowns in how 5G will achieve these speeds, it is likely that 5G will rely upon edge computing
11. 01:09to reach the targets being set. Edge computing can solve the problems in three major 5G scenarios.
12. 01:18First, edge computing devices will provide connectivity and protection services for emerging and existing network devices.
13. 01:28Second, also 5G will provide cloud based applications with better connectivity and a lower latency.
14. 01:37Data processing and storage costs are still very high.
15. 01:42The hybrid solution of edge computing and 5G will reduce these costs.
16. 01:48Finally, edge computing can allow more applications to run on the edge and device reduced the delay caused by the data transmission
17. 01:57and bandwidth limitations, perform preliminary local data analysis and share part of the work for the cloud. 5G and edge computing
18. 02:09are complementary to a certain extent. On the one hand, 5G needs edge computing to drive service demnd. On the other hand
19. 02:20edge computing can provide developers and environment to create 5G applications that do not exist before.
20. 02:30EdgeAI has a wide range of application prospects in all works of life due to its high characteristics such as low traffic
21. 02:41occupancy, low latency and strong privacy.
22. 02:46It helps companies to secure their operations products to ensure high productivity, accuracy and safety.
23. 02:54Next we will look at the most potential application areas for edgeAI's
24. 02:59future development.
25. 03:04The smartphone is the current generation of intelligent mobile terminals and it is still the most important platform for
26. 03:12edgeAI applications because there is a large number of mobile apps that have reached AI functional requirements.
27. 03:21For example, in the previous video we introduced the Google's first Federated Learning application was developed for the
28. 03:29application scenarios on android phones.
29. 03:33Another examples, Siri or google assistant are also good example of edgeAI on smartphones because this technology drives
30. 03:44their voice user interface.
31. 03:46AI on mobile phones enable data processing on the device side.
32. 03:51So there's no need to deliver device data to the cloud.
33. 03:56This helps to protect the privacy and reduce the traffic.
34. 04:01Many people now believe that VR and AR glasses will become the next generation of the smart mobile terminals. At present
35. 04:11its technology is still being further improved but we can already see that it has many application requirements for more
36. 04:20efficient edgeAI
37. 04:21such as the visual recognition rendering, 3D reconstruction, video transmission and so on and so forth.
38. 04:31With the popularity of internet of things,
39. 04:34home life will introduce more and more smart applications
40. 04:41such as smart lightning control, smart kitchens etcetera.
41. 04:45These applications require the deployment of a large number of sensors and controllers in the home environment. In order to
42. 04:55protect the privacy of a household data,
43. 04:59data processing mass rely on the edge side computing so that most computing resources are restricted to the internal gateway
44. 05:09of the home.
45. 05:09And sensitive data is avoided from floating out, optimize the indoor positioning and home intrusion detection through the
46. 05:21edgeAI to obtain higher accuracy and lower latency than cloud computing.
47. 05:27For example, smart speakers such as Amazon Echo and Google Home are widely used now. They receive user instructions and response
48. 05:37through the interaction with the 3rd Party services.
49. 05:41However, smart speakers rely on the cloud for speech recognition and natural language understanding which will cause the
50. 05:50privacy of household data to be leaked.
51. 05:55So the application of edgeAI can effectively solve these problems.
52. 06:00Similarly, home entertainment will also benefit from edgeAI.
53. 06:06The system can recommend personal life service on its own allowing users to have a better entertainment experience. Sensors
54. 06:16in smart building can detect whether there are people in the room to automatically adjust the temperature, lightning when
55. 06:25the room is not in use, thereby saving costs and protecting our environment.
56. 06:32The data and analysis results collected by a smart building may change the way infrastructure is managed. For example,
57. 06:40Siemens building an automation system, integrated all building system into a single platform that can be intuitively
58. 06:51operated. Better managing functions such as fire protection, heating, lighting and video surveillance etcetera.
59. 07:01Similar concepts and technology can be further applied in the city area such as campuses or ports. In these scenarios
60. 07:10edgeAI will give full play to its core technical advantages.
61. 07:18As the scale of cities continues to expand and the data presents the characteristics of the geographic distribution,
62. 07:27edgeAI is needed to solve more task with low latency and high efficiency. For example the building of emerging warning system
63. 07:38in the city, when danger occurs such as when children falling into a swimming pool or fire occurred,
64. 07:45we expect the system to provide accurate and fast early warning.
65. 07:51In addition AI is used to coordinate and dispatch urban infrastructure and apply it to a public safety, health care and transportation
66. 08:03and many other fields to achieve the best use of resources in the entire city.
67. 08:09For example, concerning traffic jam.
68. 08:12Singapore has responses to enormous traffic challenges through connected transportation solutions.
69. 08:22It's intelligent transportation system ITS
70. 08:25supports a dynamic electronic toll collection system at the traffic volume increased sharply,
71. 08:35the highway toll rates also increases. ITS
72. 08:40also allows real time traffic information through the GPS
73. 08:44enabled taxis.
74. 08:47It shows the dynamic planning and scheduling of the public transportation structures and can make busses more accurate
75. 08:56in their system.
76. 09:00Farming and agriculture will also benefit from the edgeAI.
77. 09:06In addition to robotic arms and drones, there are also applications of precision agriculture using the internet of things
78. 09:15based on the edge computing.
79. 09:17For example the yield and Australian based agriculture technology company use sensors, data and AI measures to help farmers
80. 09:28make informed decisions about the weather, soil and plant conditions demonstrating the huge potential of
81. 09:37edgeAI in the smart farming. In the industrial manufacturing,
82. 09:43AI and edge applications are expected to play an increasingly important role in the developing smart factories driven by the
83. 09:54industry for all standards.
84. 09:56The next generation of smart factory will apply advanced robotics and machine learning technologies to software services
85. 10:04and industrial internet of things to increase protection capacity and maximize efficiency.
86. 10:12Edge computing and AI used local sensors to control and manage output this significantly improving efficiency and reducing
87. 10:22the errors.
88. 10:23For instance, the edge system can response to input within milliseconds either making adjustments to solve the issue of shutting
89. 10:32down the production line to prevent a serious problem.
90. 10:37Healthcare has entered the era of data.
91. 10:41The popularity of health awareness has made medical advance devices and terminals more and more diversified and edgeAI will
92. 10:53help facilitate and promote the patient care and enhance the operational efficiency.
93. 11:00The first is the pre hospital first aid.
94. 11:03So, when the emerging patients arrive at the hospital or transfer between two hospitals, most of the current emerging medical
95. 11:12services are deployed in the cloud which is easy affected and restricted by the mobile environment, extreme wealth and extreme
96. 11:22weather.
97. 11:23EdgeAI can establish a bidirectional communication channels between the ambulance and the hospital, realize the real time image
98. 11:35and language processing and improve timeless and efficiency.
99. 11:41The second is smart variable devices. The current smart variable devices affected by computing power and only play
100. 11:51a role in the data collection.
101. 11:53However, this is an active research field such as deploying AI models on viable devices to monitor human
102. 12:02emotions in the future.
103. 12:05More lightweight AI algorithms will be deployed on the devices to monitor, analyze and predict the health data in real time
104. 12:14helping humans better understand their physical conditions.
105. 12:20I only list the part of the potential application scenarios of
106. 12:24edgeAI but what is certain is that the era of edgeAI
107. 12:28is coming and if you change the AI model, the ecology of AI computing, software hardware and make our lives more comfortable
108. 12:38and intelligent.
109. 12:42In this week's practical session we will experience collaborative influence for image classification with Sedna.
110. 12:50The time required is about 3-6 hours.
111. 12:53I hope you all have fun and have great success.
112. 13:00Thank you for watching the video.

To enable the transcript, please select a language in the video player settings menu.