2018 Global Artificial Intelligence Breakthrough Technology TOP10

Artificial intelligence is a high-tech, wide-area, multi-dimensional, interdisciplinary master. From the pure computer application based on big data and around the Internet, it is gradually derived from all aspects of people's daily production and life, and improves and changes us in the nuances. At present, many new technologies and new models have been gradually put into practical use, but most of them are still in the stage of promotion, experimentation and research. How to grasp the great opportunities of promoting artificial intelligence technology, so that the wider people can use mobile phones like mobile phones. With artificial intelligence, this is a must-answer for the development of the era that our generation must face.

In 2018, artificial intelligence technology has made breakthroughs in many aspects. Technology companies at home and abroad are constantly trying to apply artificial intelligence to more fields. Regardless of technology giants or start-ups, they are committed to continuous innovation and promote technological advancement. Let's take a look at ten breakthrough technologies in the field of artificial intelligence in China and foreign countries.

1. Neural network based machine translation

Reason for selection: Translation is the most important branch of “natural language processing” and a difficult one. In the early years, machine translation was also ridiculed as “low-level translation.” Nowadays, the machine translation accuracy of neural networks is greatly improved, comparable to professional human translation. Our well-known Google Translate, Microsoft Voice Translator, and Sogou Speech Recognition are all based on this technology.

Technological breakthrough: Machine translation is a research field that researchers have been attacking for decades. Its core technology is a deep neural network with massive nodes, which can automatically learn translation knowledge from the corpus.

The process of the human brain's processing of language is undoubtedly one of the most complex cognitive processes. Many people once thought that machine translation could not reach the level of human translation. Multilayer sensing in neural networks attempts to simulate the process of multi-layer delivery of human brain nerves, but usually no more than three layers. In 2006, scientists proposed a deep learning algorithm for neural networks, enabling neural network training with at least 7 layers. Because it can better simulate the process of multi-layer deep transmission of human brain neurons, it has a very obvious breakthrough performance in solving some complex problems.

In March of this year, Microsoft announced that its machine translation system was first developed in the Chinese-English translation of General News, achieving another milestone in natural language processing, and surpassing the time of machine translation beyond human amateur translators. It took a full seven years.

Significant significance: Machine translation based on neural network directly improves the effect of literal translation before word, and upgrades to translate in whole sentences.

Comments: Machine translation in hand, the whole sea is not a problem.

2. Robot autonomous navigation based on multi-sensor cross-border fusion

Reason for selection: The ultimate goal of robots is to provide intelligent services for human beings. Autonomous navigation is a technical barrier that humans have been trying to overcome in recent years. PowerVision integrates underwater sensors through so-called multi-sensors such as sonar and vision. It can realize autonomous navigation and intelligent recognition, and has made breakthroughs in the field of intelligent robots.

Technological breakthroughs: As the application scenarios and tasks of robots become more complex, it is difficult for a single sensor to meet application requirements. Multi-sensor information fusion is highly dependent on hardware resources. PowerVision is based on various sensors, inertial navigation, motion control, camera, visual inspection/recognition, SLAM and other cores accumulated in the robot industry for many years. The in-depth study of technology and deep learning, through the system architecture and optimization design of the integrated platform of the embedded end, breaks through the limitations of the hardware resources of the mobile platform, so that the underwater robot can more accurately, intelligently and comprehensively perceive the target, and has the right The ability of underwater targets to lock, detect, identify, and follow.

Significant significance: through the combination of navigation control, sonar, visual technology and deep learning, the robot's positioning and autonomous navigation in the global environment, as well as intelligent applications, provide powerful technology for human exploration of broader underwater applications. Guarantee.

Comments: This may be the best way to land artificial intelligence in the future.

3. DuerOS conversational artificial intelligence system

Reason for selection: DuerOS3.0 can bring users an epoch-making natural dialogue interaction, including the comprehensive upgrade of natural language interaction technology including emotional voice broadcast and voiceprint recognition.

Technical breakthrough: DuerOS is a conversational AI operating system developed by Baidu's Secret Division. It has more than 250 skills in 10 categories. DuerOS includes a complete interactive process from speech recognition to voice announcements to on-screen display, as well as core technologies behind natural language understanding, dialog state control, natural language generation, search, etc. that support interactions. These technologies support the application layer and capabilities. Layer implementation.

On July 4, 2018, the latest DuerOS 3.0 was officially released, enabling the enabling products to implement multi-round speech error correction, complex progressive intent recognition and logical conditional intent recognition, thus more accurately determining user intent, and ultimately Achieve functional upgrades - use extended features to understand user behavior. Based on this, DuerOS 3.0 provides more than 20 cross-scenario, cross-device solutions including screen device solutions, Bluetooth device solutions and industry solutions.

Significant significance: DuerOS is the first to open commercialization of the AI ??era, and will provide complete application program support for eco-partners in terms of products, content and promotion, and accelerate the landing of AI equipment.

Comments: Smallness, what do I write about the next technology?

4. Mobile AR technology

Reason for selection: In the future, AR and AI need to be mutually blessed, and AR can be compared to the eyes of AI.

Technological breakthrough: integrated many computer technology and graphic imaging technologies, including real-time rendering technology, spatial location tracking, image recognition, human-computer interaction, display technology, cloud storage, data transmission, content development tools and other fields.

The AR technology not only displays the information of the real world, but also displays the virtual information at the same time, and the two kinds of information complement each other and superimpose. In visual augmented reality, users can use the head-mounted display to combine the real world with computer graphics to see the real world around it. Augmented reality technology includes new technologies and new methods such as multimedia, 3D modeling, real-time video display and control, multi-sensor fusion, real-time tracking and registration, scene fusion, etc., providing a new way for humans to perceive information.

Future mobile AR technologies will evolve to creative AR applications, location-based AR experiences, and multi-person AR experiences.

Significant significance: With the release of Apple ARKit and Google ARCore, mobile AR is significant on both mobile platforms. This also means that 500 million AR-capable mobile devices are attracting all companies, and these companies are combining data with APIs to create new AR experiences for users.

Comments: I feel that I am not far from entering the real world of the second dimension.

5. Biometrics

Reason for selection: Behavior recognition technology is applied to security, and a lock is added for security.

Technical breakthrough: It is a very common behavior to install cameras in stores. However, ordinary cameras can only record the behavior of people in the store. If it is found that theft needs to be manually checked by watching the monitoring records, it takes time and effort.

Recently, the Japanese telecommunications giant announced that it has developed a new type of artificial intelligence security camera called "AI Guardman". This camera can accurately predict the human action intentions before the theft, so as to help the store. Identify theft and discover potential store pickpockets.

This artificial intelligence system uses open source technology to scan video streams in real time and predict people's posture. When there is a behavior in the monitoring, the system will try to match the posture data with the predefined "suspicious" behavior, and once found, will notify the owner through the relevant mobile app. According to relevant media reports, this product has reduced the store's theft by about 40%.

Significant significance: Through behavioral recognition technology, criminals can be discovered in the first time, criminal behavior can be predicted, and store security can be effectively protected.

Comments: Going to a store in Japan must be decisive, and it is not good to be misunderstood as a thief.

6. Robot process automation

Reason for selection: Robotic process automation can help or even replace humans with a large number of simple, single, repetitive and arduous tasks, and is more efficient and zero-error.

Technological breakthroughs: Robotic Process Automation (RPA) is the efficiency of automation and work through the use of high-performance cognitive technologies. Humans can write a variety of services, such as browsers, clouds, and various software, by writing workflows that need to be done manually on the operator interface.

According to Gartner data, in the past year, 300 large-scale commercial giants in the world have successively launched RPA projects to automate the original manual process. As technology advances, RPA will incorporate more artificial intelligence technologies, namely Intelligent Process Automation. It is equivalent to adding reasoning, judgment, and decision-making skills based on deep learning and cognitive technology on the basis of rule-based automation foundation (RPA).

Significant significance: Robotic process automation can greatly improve the efficiency of the company, reduce personnel input, help companies reduce costs, let humans release productivity, and free up hands to do higher value things.

Comments: The little white-collar worker was shivering and shivering, please don't fire me.

7. PixelPlayer, pixel level sound source positioning system

Reason for selection: Separation of sound sources from the perspective of visual and auditory signals, opening up new avenues for research.

Technical breakthroughs: When editing music, the equalizer is used to tune out the bass in the music, and researchers at the Massachusetts Institute of Technology's Computer Science and Artificial Intelligence Labs have found better solutions. The new system PixelPlayer they developed.

PixelPlayer combines sound and image information, and the machine learning system can unrecognically identify objects from images or sounds, locate targets in images, and separate sounds produced by targets. When we give an input video, PixelPlayer can jointly separate the dubbing into the target component and locate the target component on the image frame. PixelPlayer allows you to locate the sound source on each pixel of the video.

To put it simply, PixelPlayer recognizes which object in the video is emitting sound and separates the sound.

Significant significance: PixelPlayer is able to filter accompaniment and recognize sound sources, not only to help humans deal with music, but also to help robots better understand the environmental sounds produced by other objects.

Comments: If the teacher uses this system, he can immediately find out the classmates who are in the classroom.

8. Deep learning chip that combines high-precision learning with low-precision reasoning

Reason for selection: This chip covers the best training and best reasoning currently known to keep all processor components available for data and work.

Technology breakthrough: The deep learning chip is one of the projects IBM is working on. IBM set the target utilization rate of this chip at 90%. This will be a qualitative breakthrough, and in order to achieve this breakthrough, IBM's R&D team has made two innovations.

First, the low utilization rate is usually due to the data flow bottleneck that exists around the chip. To overcome these information barriers, the project's team developed a "customized" data flow system. The data stream system is a network solution that accelerates the transfer of data from one processing engine to the next. It is also optimized for whether it is a learning task or a reasoning task and different precision.

The second innovation is that the team uses a specially designed "scratchpad" form of on-chip memory instead of the traditional cache on the CPU or GPU. The goal of building it is to keep the data flowing through the chip's processing engine and ensure that the data is in the right place at the right time. In order to achieve 90% utilization, IBM must make the designed scratchpad have a huge read/write bandwidth.

Significant: The chip can perform all three major deep learning AIs today: Convolutional Neural Network (CNN), Multilayer Perceptron (MLP), and Long-Short Term Memory (LSTM). These technologies collectively dominate language, visual, and natural language processing.

Comments: At present, deep learning technology is in the bottleneck, and IBM's deep learning chip may be able to reverse this situation.

9. Intelligent agent training platform

Reason for selection: Intelligent agent technology makes computer applications tend to be humanized and personalized. It is a mobile computing entity that can perform a set of operations in an active service manner without the need for immediate human intervention.

Technological breakthroughs: Intelligent agent technology has the rich knowledge, strategies and related data needed to solve problems, and can carry out related reasoning and intelligent calculation. Intelligent agents can also infer user's intentions and interests when users do not give very clear requirements. Or hobbies, and in the best way to complete the task for them, and can automatically reject some unreasonable or potentially harmful requirements; intelligent agents also learn from their own self-learning, can adjust their behavior according to the environment, thereby improving The ability to handle problems.

Intelligent agent technology can be applied to business, intelligent search agents, digital libraries, e-commerce and distance education research, and now it is also used in the game field.

Unity is one of the world's leading game development companies. Last year it launched the machine learning platform ML-Agents, allowing AI developers and researchers to use evolutionary strategies, deep reinforcement learning and other training methods to train in Unity simulation and gaming environments. Intelligent agent. This method of simulation training is widely used in industrial robots, drones, unmanned vehicles and game character design.

Significant: Unity's mission is to enable machine learning researchers to get the most powerful training scenarios, providing a research platform for the fast-growing AI enthusiast community to explore deep learning.

Comments: NPC is getting smarter and may not be a good thing for the game hand.

10. In-ear artificial intelligence

Reason for selection: The headset has changed from a smart device to a smart wearable device, which can extend the usage time and scene indefinitely.

Technical breakthrough: Apple has added Siri evoke, built-in chips, etc. in AirPods 2, which will be launched this year. It can collect walking steps, heart rate data and body temperature, etc. It can also capture the movement of the user's head or even the positioning position through the built-in gyro. Of course, you can also accept commands through the microphone and then feedback through the speakers.

Google's real-time translation wireless headset PixelBuds is often used to compare with AirPods. PixelBuds evokes Google Assistant very quickly. Just put your finger on the headset's touchpad and talk to Google Assistant at almost the same time.

AirPods and PixelBuds allow us to redefine the role of headphones for humans. Compared to smart watches on the wrist, headphones can obviously make voice interaction more convenient, and do not need to take up valuable visual space when receiving information. It also turns smart speaker-style far-field interactions into more natural and faster near-field interactions.

Significant significance: smart headphones make the assistants more close to accompanying, Apple, Google's involvement, will certainly bring a wave of AI in the ears.

Comments: Now our mobile phone does not leave, and the headset may not leave the ear in the future.

We see these technology companies or research teams learning translation, research fishing, developing games, doing music... It seems to be “not doing business”, and this just proves that artificial intelligence is not just a technology or a product.

From IBM, Apple, to Google, Baidu, all the artificial intelligence giants are trying to connect software, hardware, and application scenarios. Smart technology companies no longer focus on their traditional business, but focus on the future, constantly innovating technology, and creating an integrated ecosystem through cross-border integration.