Alibaba has recently exploded, and Alibaba's ai model has aroused everyone's attention in the world record of SQuAD, the top event in reading comprehension. Follow the small series to find out what it is, see the future development trend of Alibaba artificial intelligence and What is the future trend of global artificial intelligence?
Alibaba's ai model broke the world recordA few days ago, Stanford, the top event in the machine reading comprehension field initiated by Stanford University, has a new record. Alibaba has broken the world record with 82.440 accuracy and surpassed the human 82.304.
Pranav Rajpurkar, head of SQuAD, said that the first model (SLQA + submitted by the Alibaba iDST team) surpassed human performance in precision matching. The next challenge: fuzzy matching, humans still lead by 2.5 points.
It is understood that the SQuAD competition has built a large-scale machine reading comprehension data set (including 100,000 questions). The article is based on more than 500 Wikipedia articles. It aims to use this set of questions to sort out the clues and see if the machine learning model can Give the exact answer to the question after a lot of information processing.
Siro, chief scientist of natural language processing at the Alibaba Institute, said in an announcement that the accuracy of the answers given by the machine would be high for objective questions such as "Why is it raining?" According to the announcement, the technology can be gradually applied to a wide range of practical applications such as customer service, museum guides, online answering patient medical issues, etc., thereby reducing the need for manpower input in an unprecedented way.
The major breakthrough in this technology stems from the deep neural network model based on the layered fusion attention mechanism proposed by the Alibaba research team. The model can simulate some behaviors of human beings when doing reading comprehension problems, including combining chapter content examination questions, reading articles repeatedly with questions, avoiding forgetting and observing in reading.
It is understood that this technology has been widely used inside Alibaba. For example, every year, 11 people will have a large number of customers to consult the rules of the event. Ali Xiaomi team is the most natural way to interact by using the technology of the Siro team to let the machine read the rules directly and provide users with rules interpretation services.
For example, customers also ask a large number of basic questions about individual products, and these questions are actually answered on the product details page. Now through machine reading comprehension technology, the machine can read and answer the product description text in the details page more intelligently, reducing the cost of service and increasing the purchase conversion rate.
According to Xiaobian, the natural language processing team led by Siro supports the technical needs of Alibaba's entire ecosystem. The AliNLP natural language technology platform developed by them is called 120 billion+ times a day, and the Alitranx translation system provides more than 700 million+ calls per day for 20 languages. Previously, he achieved the first place in the world in the 2016 ACM CIKM personalized e-commerce search, the 2017 IJCNLP Chinese grammar test CGED evaluation, and the 2017 US Standards and Metrology Bureau TAC evaluation English entity classification.
Ali's layout in artificial intelligence mainly has two directions: one is the integration of e-commerce and merchants, and the other is to give technical support to manufacturers.
The Alibaba Artificial Intelligence Laboratory was unveiled on July 5, 2017, focusing on consumer-grade AI products. The first product is the familiar smart voice terminal device "Tmall Elf X1".
iDST (Data Science and Technology Research Institute) is known as Alibaba's most mysterious research institution, distributed in Hangzhou, Beijing, Seattle, Silicon Valley, etc. It is the core team of Alibaba's research and development of artificial intelligence technology, Alibaba's NASA program. Artificial intelligence brain.
Xiaobian learned that in addition to the artificial intelligence laboratory, the Institute of Data Science and Technology iDST, Ali also has the Ali Institute, VR Lab, and Ant Financial also has its own artificial intelligence team. In addition, in March 2017, Ali announced the launch of the "NASA" program to form new teams and establish new mechanisms and methods for core technologies such as machine learning, chips, IoT systems, and biometrics.
Extended reading: the development trend of artificial intelligenceTrend 1: Big companies will benefit from artificial intelligence
Amazon, Google, Facebook and IBM will lead the way in artificial intelligence. As big companies, they have the right resources to collect data, so there is more data to use and how these giants are laid out.
Amazon: Investing in artificial intelligence for more than 20 years, grabbing more than 5B of web page data, more than 500,000 JPEG images and corresponding JSON metadata for use in Amazon's operations center products. More than 250 million data is captured every day in the world of radio, magazines and online news. Nearly 100M images and videos are captured every day with audio and visual features and annotations. Amazon Echo series speakers have occupied more than 70% of the voice assistant market.
Google: With the world's largest database, focus on application and product development, not long-term AI research. GoogleBrain has a team of more than 1,300 researchers and has a 23.8% share of the voice assistant market. Machine learning using the TensorFlow open source platform, allowing anyone to access the machine learning platform. The size of the Google Earth database is estimated to be 3017TB or about 3PB, and Google StreetView has about 20PB of Street View photos.
Google is likely to be at the forefront of application and product development and service deployment. It is not only the first company to start researching artificial intelligence, but also has 70,000 employees. In addition, Google has a deep learning artificial intelligence research project, GoogleBrain, which has a team with its own research agenda covering machine learning, natural language understanding, machine learning algorithms and techniques, and robotics.
Facebook: Working with 2.5B of content and more than 500 terabytes of data per day, Facebook ArTIcialIntelligence Researchers (FAIR) has about 80 researchers and engineers, generating 2 billion "likes" and 30 million photos a day, scanning about 105 terabytes of data every 30 minutes. There is a 62,000 square foot data center that can accommodate up to 500 racks. Translate 2 billion user posts in more than 40 languages ​​every day, and 80 million users use these translations every day.
IBM: A 10-year, $240 million investment is planned to create the MIT-IBM Watson Artificial Intelligence Lab. With more than 2,000 AI employees worldwide, more than 600 AI employees at IBM headquarters, and Watson users across six continents and more than 25 countries, IBM has invested $1 billion in the Watson project, including $100 million in venture capital. More than 7,000 applications have been created through the Watson ecosystem.
Trend 2: Integration of algorithms and technologies
All Tier 2 capital companies investing in artificial intelligence, such as Intel, Salesforce, and Twitter, will follow the big companies that own the data and use their data algorithms and artificial intelligence. Data transactions will occur between industry participants and it is likely that algorithms and techniques will be integrated. The trading of data and the integration of algorithms and technologies will make artificial intelligence even more important.
As larger companies such as Google and Facebook acquire smaller companies, more algorithms will be integrated into their core platforms or solutions. DeepMind, an artificial intelligence company based in London, UK, built a universal learning algorithm that was acquired by Google to gain a commercial advantage over other technology companies. On the other hand, Facebook acquired Wit.ai to enhance its speech recognition and voice interface. The company also acquired the artificial intelligence startup Ozlo to complement its M virtual assistant service.
Trend 3: Data Crowdsourcing
All artificial intelligence companies are pursuing huge databases to achieve their ambitions for artificial intelligence. These companies will begin to get a lot of data through crowdsourcing. Companies have found a way to assess the quality and authenticity of crowdsourced data, not only to provide convenience to the business, but also to feed back information to consumers.
Joel Gurin, founder and editor of OpenDataNow.com, said, “We live in a crowded culture environment where more and more people are willing and interested to share what they know through social media.â€
Through crowdsourcing, Google has acquired a large number of images and built imaging algorithms. The company also uses crowdsourcing to help improve services such as translation, transcription, handwriting recognition and map applications. Amazon also used crowdsourcing technology to improve Alexa's 15,000 existing skills.
Trend 4: More mergers will occur
CBInsights statistics show that AI's acquisition competition has begun. 2018 will be the year we can see the most acquisitions and acquisitions of companies, because these companies must compete for intellectual capital and talent will not be eliminated. All small companies of machine learning/artificial intelligence will be acquired by large companies. There are two reasons:
AI can't work without the help of a database. Because large companies have a large database, they will put tremendous pressure on small companies. Without the support of the database, the algorithm will be useless. Also, if there is no algorithm, the data is almost useless. Data is at the heart of the algorithm, and a large amount of data is critical. Hod Lipson, a robotic engineer and director at the Creative Robotics Laboratory at Columbia University, said, "Data is fuel, algorithms are engines."
Trend 5: Open democratization tools will gain market share
Large companies will begin to open their algorithms and other tools to gain market share. Market-based barriers to entry of data and algorithms will decrease, and new applications for artificial intelligence will increase. Through open platforms and democratization, small companies that cannot use artificial intelligence tools will have access to a large amount of data to study artificial intelligence algorithms.
As Google CEO Sundar Picha said when it comes to democratized artificial intelligence, “One of the most exciting things we can do is to make machine learning and artificial intelligence It's no longer so mysterious. It's important to make it accessible to everyone." In addition, frameworks, SDKsandAPIs will become the standard for all major vendors to open to consumers. All companies will adopt the SaaS & PaaS business model.
Trend 6: Human-computer interaction will be improved
Siri and Alexa are probably the most popular human-computer interaction tools, and more robot-based solutions will be the threshold for artificial intelligence companies to enter the industry. For example, although the machine has been programmed for speech analysis and facial recognition, the machine has to recognize your emotions based on your voice, that is, to perform emotional analysis.
Manufacturing automation and non-consumer focus solutions will be the first solution/application to be improved. Manufacturing automation will be largely due to labor cost savings, including automation, robotics and advanced manufacturing techniques. Improvements in non-consumer solutions, such as human-computer interactions that perform tasks in agriculture and medicine, will also become popular in 2018.
Trend 7: Artificial intelligence will gradually affect all vertical areas
Manufacturing, customer service, healthcare, healthcare and transportation have been affected by AI, and autonomous vehicles are expected to be available in 2018. Next year, there will be more areas affected by artificial intelligence. The following are examples of the impact of artificial intelligence on different industries:
Insurance - AI will improve the claims process through automation.
Law - NLP can summarise thousands of pages of legal documents in minutes, reducing inspection time and efficiency.
PR&media - AI will help process data quickly.
Education - the development of virtual tutors; artificial intelligence to help scores; develop adaptive learning plans, games and software; AI-oriented personalized education programs will change the interaction between students and teachers.
Health – Machine learning can be used to create more sophisticated and accurate ways to predict disease before a patient has symptoms
Just as the industrial revolution changed everything almost 100 years ago, artificial intelligence will change the world in the next few years.
Trend 8: Security, privacy and ethics issues
Under the umbrella of artificial intelligence, issues such as machine learning and big data are easy to reach security and privacy issues. Sometimes infrastructure plays a very important role. Security requirements related to privacy issues, such as keeping bank accounts and health information confidential, will place greater demands on the security of research. In 2018, issues related to security and privacy will be resolved, and this year is also a year in which new developments in artificial intelligence may occur.
The ethical issues of artificial intelligence will also become the main problem in 2018. The ethical and moral issues that need to be solved include the benefits and disadvantages of artificial intelligence for human beings. People are also worried about the possibility of robots replacing humans, such as nurses, therapists or police, and another problem that needs to be dealt with is autonomous weapons.
24V Battery Pack ,Large Battery Pack,24 V Battery Pack,24V Lithium Ion Battery Pack
Zhejiang Casnovo Materials Co., Ltd. , https://www.casnovonewenergy.com