 |
| Dr Min addresses the crowd at his keynote during the NVIDIA AI Conference. |
Dr Wanli Min, Chief AI Scientist, Alibaba Group, says the smart city is something tangible that can be achieved in the same way as combining atoms into a functional molecule, only it is about smart technologies and smart people acting in coordination.
"Aim for system-wide intelligence in order to maximise returns," he said, speaking during a keynote during the first NVIDIA AI Conference in Singapore.
 |
| Alibaba's ET Brain harnesses various technologies for smarter outcomes. PAI is Alibaba's machine learning platform; OCR stands for optical character recognition, or reading text from an image. |
DI, what Alibaba calls "data intelligence", and ET, what Alibaba calls "extreme tech", push the envelope on what is possible with artificial intelligence (AI). With data from mobile apps and other sensors sending feeds in real time combined with publicly available data, it is possible to create actionable insights on-the-fly in the cloud and change things in real-time, Dr Min said. For example, on-demand bus services can be generated instead of having them adhere to fixed schedules. The same ET can be applied to event detection, traffic recognition, face recognition, and so on. Use cases include:
- Taking over human stenography (note-taking) in digital court proceedings
- Intelligent conversations at a digital call centre. This technology is currently supporting the Alipay customer service department. "More than 90% of calls are taken by this technology," Dr Min said.
- Voice identification, distinguishing between situations where there may be violence, comparing a voice in real-time to a list of 10 million samples in whitelists or blacklists
 |
| Intelligent image recognition in e-commerce. |
- Intelligent image recognition in e-commerce - users can provide a picture of a product to locate merchants in an e-marketplace which are selling the product. The search works even if the object is blurred, displayed at an unusual angle, or if there are other items in the picture.
"We can search on Tmall, Taobao in real time, a billion pictures in high definition, it is literally a brute force search that in the past was not possible. These days, with deep learning, it's pretty fast and it's pretty accurate," Dr Min said, adding that personalised recommendations based on previous purchases are also included in the system.
"We require additional technology for object segmentation," he added. "Every step requires huge computational power."
 |
| Image recognition powered by deep learning undergoes many steps. Object segmentation has to occur for the sample picture of a shoe to determine that the items in the background are not what the shopper wants to buy. |
"We can search on Tmall, Taobao in real time, a billion pictures in high definition, it is literally a brute force search that in the past was not possible. These days, with deep learning, it's pretty fast and it's pretty accurate," Dr Min said, adding that personalised recommendations based on previous purchases are also included in the system.
"We require additional technology for object segmentation," he added. "Every step requires huge computational power."
 |
| AI can help online merchants with automated tags and signage. |
- Intelligent tagging of pictures for e-commerce merchants. The system will describe the picture and
understands the intent of the image, interpreting a picture of a child wearing a hat carrying a box as an image advertising Christmas gifts for children, for example. The system can also look at a selection of sentences and enlarge the most important phrases while centralising them. "It decides what is to be highlighted, and which are similar messages (to ignore)," Dr Min said.
- Intelligent traffic sensing. AI systems can conduct real-time video analyses on many streams of raw video at the same time to identify traffic problems. Incidents can be identified and highlighted, right down to tracking offenders through the colour of the car, the slope of the windshield, and their licence plates. The system has been able to identify hit and run drivers and intercept them in real time.
In Hangzhou, China, where Alibaba is headquartered, a "city brain" analyses traffic cameras and aid traffic enforcers in real-time responses. The system can identify illegal behaviours and adjust traffic signals in real time, cutting down emergency response arrival time by 50%.
"It's about how automatically, how fast you can do it in real time," Dr Min said.
The AI system can manipulate traffic signals based on the route that an ambulance is taking to create an “intelligent green wave”. With the system, the ambulance will experience a green signal at any intersection.
- Intelligent monitoring of plants in a fixed location over time - fixed cameras take pictures of a cornfield over time to detect plant growth patterns, eventually determining whether the plants are healthy, if there are any pests, and predicting production yields. "We are doing that now in Suzhou in real time," Dr Min said.
understands the intent of the image, interpreting a picture of a child wearing a hat carrying a box as an image advertising Christmas gifts for children, for example. The system can also look at a selection of sentences and enlarge the most important phrases while centralising them. "It decides what is to be highlighted, and which are similar messages (to ignore)," Dr Min said.
 |
| Real-time video analysis for tracking hit and run drivers. |
 |
| Creating a green wave for emergency response vehicles, cutting down travel times. |
- Intelligent traffic sensing. AI systems can conduct real-time video analyses on many streams of raw video at the same time to identify traffic problems. Incidents can be identified and highlighted, right down to tracking offenders through the colour of the car, the slope of the windshield, and their licence plates. The system has been able to identify hit and run drivers and intercept them in real time.
In Hangzhou, China, where Alibaba is headquartered, a "city brain" analyses traffic cameras and aid traffic enforcers in real-time responses. The system can identify illegal behaviours and adjust traffic signals in real time, cutting down emergency response arrival time by 50%.
"It's about how automatically, how fast you can do it in real time," Dr Min said.
The AI system can manipulate traffic signals based on the route that an ambulance is taking to create an “intelligent green wave”. With the system, the ambulance will experience a green signal at any intersection.
Logically, such a system can also be used to check machinery for defects that occur over time, without requiring that the equipment is out of service during the inspection. A system used in the Chinese railways does exactly that, Dr Min shared. "In the past we needed 10,000 experienced engineers to inspect defects overnight. With this technology, we have cut (manpower) by 90%."
Fujian province provides a related use case in real-time water pollution prediction. Remote sensing data and Internet of Things (IoT) data are coupled to generate a map that predicts the index of water pollution and how does pollution will propagate along a river, Dr Min shared.
 |
| Integrated fishery analysis using AI. |
Similar technology is in use to determine current fish populations, which have a bearing on fishing rights. Historical data is useless here, and fish are mobile whereas parts of the sea can be protected. AI can determine which parts have been overfished and need to be protected. "This requires a lot of computational power," Dr Min said.
"To enable data intelligence (we) need hardware. We have had a long collaboration with NVIDIA and Alibaba Cloud to generate actionable insights," he said.
"Ten years ago we didn't have enough computation power," Dr Min concluded. "We believe smart cities will become a reality and if we work in that direction, if you believe it you will see it. The city dream is just the starting point. This time it's not about IT, it's about DT, data transformation. Just do it until you hit a roadblock and think about the implications. At this stage our limitations are bounded by our concerns of privacy. Don't let these concerns bound you. Believe and you will see."
Explore:
Read the TechTrade Asia blog posts on:
NVIDIA's AI strategy
Cutting-edge solutions from the NVIDIA AI Conference showfloor
Singapore's approach to AI
The EDB-NVIDIA Future Talents Program
The shared NVIDIA and NSCC platform for AI initiatives in Singapore
Hashtag: #NVAICONFERENCE17
No comments:
Post a Comment