Set micro-network launched integrated circuit micro-letter public number: ' Every day IC ', major news real-time release, every day IC, every day set micro-nets, accumulate micro-Cheng! Replication Laoyaoic Micro-letter public number search adds attention.
1. Next year handset differentiation competition brand factory wins AI new battlefield;
After 2017 years of high competition in the global smartphone market, manufacturers look forward to the 2018 mobile phone market will present a new outlook, not only the mobile phone hardware specifications continue to innovate, more importantly, the 2018 mobile phone manufacturers will expand the introduction of artificial intelligence (AI) applications, enhance virtual reality (VR) and augmented Reality (AR) support, As for the mobile phone industry layout and market changes, manufacturers expect the mainland mobile phone factory forces will further expand, although the high-end mobile phone demand is expected to return to warm, but the competition will be more fierce fighting, and the related 0 component shortage will continue. In addition to continuous hardware innovation and price/performance competition in the 2018 smartphone market, the first-line handset manufacturers will still play an important role in leading the global mobile phone specification, including Samsung Electronics (Samsung Electronics) Note 8, Apple IPhone in 2017. X's applications in artificial intelligence, as well as dual lenses, innovative identification systems, high specification screens, etc., are expected to become an important direction for other competitors to follow up in 2018. Samsung's first high-end mobile phone with flexible OLED design will be unveiled at CES in 2018, and whether it triggers a new wave of mobile phone factory high-end model competition, as well as the panel supply chain and hardware terminal vendors of the competing changes, become the focus of the industry attention. It is noteworthy that, with the artificial intelligence in the application of more diverse mobile phones, 2018, including Samsung, Apple, Huawei, HTC and other mobile phone manufacturers, will be introduced artificial intelligence-related applications, as a key selling point of mobile phone differentiation, the 2018 mobile phone brand Factory will be in the new battlefield of the full victory. And with Google, Amazon, and other voice assistant technology development more mature, it is expected that the mobile phone and the Internet, related to the use of the surrounding products more enriched. As for the very popular VR in recent years, AR and other applications, is also the 2018 mobile phone development can not be ignored the key indicators, manufacturers believe that as mobile phone specifications continue to improve, 5G network into the trial operation phase, is expected to allow mobile phone with the experience of VR increase, especially in the AR aspect, the global handset brand manufacturers have intensified support strength, Within 2018 years of building depth sensor mobile phone style will be increased, coupled with mobile phone display image quality significantly improved, also conducive to the application of AR, let mobile phones in commercial, entertainment and other applications more yuan. In the mobile phone market competition, mainland mobile phone factory Huawei, Oppo, millet, etc. by the mainland market support, plus overseas market sales, manufacturers are expected to 2018 mainland China's three mobile phone brand factory annual shipments, have the opportunity to reach more than 100 million, and for the lifting of mobile phone sales profit and improve brand awareness, 2018 Mainland mobile phone manufacturers in the medium and high-end models of the market will be more strong, and more pressure on competitors. As for the 0 component supply, due to the continued increase in high-end mobile phone shipments, coupled with increased mobile phone innovation, manufacturers expect mobile phone upstream key 0 components fear of continuing to be in a tight state, in addition to panels, memory and other key 0 components, including PCB, passive components, etc. may also appear in short supply, And in the face of the 0 component price may climb, will let the handset brand manufacturer faces the profit pressure to increase, 2018 handset market development and the industry change is still big, continues to affect the supply chain operation trend. Digitimes
2.AI Chip company is force to complete the 450 million yuan a round of financing;
Sina Science and technology news December 15 night, recently, AI chip start-up team, Force announced the completion by the map technology, Yunfeng Fund, Sequoia Capital, high jianling capital of the 450 million yuan a round of financing. Three focus on different investment companies and AI enterprises in accordance with the map technology investment Portfolio, the joint focus on AI chip, can be seen in the capital of the AI chip development of the future has been unanimously recognized.
From the official website introduction, Shanghai Yi Know Electronic Technology Co., Ltd. (see FORCE) is just set up this year, the new company, from Chip design, algorithmic software, system development field of senior experts founded. Company's current scale of 11-50 people, legal representative is Xu Ru 淏. For the team-specific members, the official did not give a detailed description.
And this Zhang Zhenning on behalf of the company's external voice, currently as the force of the VP Marketing. According to the information on the Internet, Zhang Zhenning graduated from Shanghai Jiao Tong University with the background of MIT. Prior to the famous Japanese electronics company TDK and China Tencent Company.
According to the official introduction, chip virtualization technology, in need of flexible calculation of the scene doubled the chip utilization rate, such as the overall AI cloud utilization rate doubled, similar to the virtualization of the CPU to cloud computing flexibility to bring a multiplier cost savings. Through a large number of research and experiments on the chip architecture and modularity to ensure full transparent virtualization capability in cloud applications. In addition, the combination of self-developed firmware and TFDL Software SDK can achieve the calculation of all kinds of neural network model acceleration, the efficiency of the acceleration unit between 90%-95%, compared to Nvidia's mainstream computing card can achieve more than 5 times times the power and cost savings.
AI chip as the basis of the development of artificial intelligence, in the global scope has already become the most important strategic heights of the industry giants. Chip companies Nvidia and Intel, to acquire or self-research the way to expand the product boundaries; Internet companies Google and Facebook are investing in AI chip development to meet their ever-exploding computing needs; and domestic such as Ali, Tencent also unwilling to be outdone, has already begun to layout AI chip enterprises at home and abroad. The force financing can be expected, with the global vision of the top investors and the rapid accession of capital, industrial resources will be further gathered. (Sing)
3. Horizon Huang: The scale of industrial landing determines the value of artificial intelligence company;
Wen/Yang Yulin
Many people call 2017 the year of the artificial intelligence explosion, and in the past 50 weeks of 2017, AI-related content has been in almost every week's VC news.
At present, in the field of artificial intelligence entrepreneurship, there are two directions: 1, based on algorithms, do algorithm research; 2, hardware-oriented, AI chip; For the general start-up companies, most of the choice to study the algorithm alone, or to do the AI chip alone, only Google or bat-level manufacturers, will also study the algorithm, and the development of chips.
and Horizon Technology is a very special start-up company, not only does the algorithm also produce its own chips, the horizon hopes to put the ' brain ' on more than 1,000 devices, using the algorithm to integrate algorithms into High-performance, Low-power, low-cost embedded AI processors and hardware and software platforms, so that they have a sense, interaction, Understand the intelligence of decision making. At present, the horizon with the domestic and international top car Tier 1, OEMs and toys, home appliance manufacturers have launched a deep cooperation, has successfully introduced mass production products.
At the 2017 Global Youth entrepreneur Conference, Tencent Technology interviewed Dr. Huang, the co-founder of the horizon, and chatted with him about the 2017 years of change in the AI industry and the opportunities that might emerge in the future.
The following is an interview record, Tencent Science and Technology collation release:
The hardware is to better fit the algorithm, 2018 start to try to mass production of chips
2015, when Huang resigned as the director architect of the depth study Institute of Baidu, and when he was the deputy dean of the depth Study Institute of Baidu Yukei and other people to create a horizon robot, dedicated to providing High-performance, Low-power, low-cost, fully open embedded artificial intelligence solutions.
In the horizon planning, not only to do algorithms, but also to do artificial intelligence chip, such a practice is not understood at that time, as a start-up if the research algorithm and hardware is a waste of resources, while in the chip area although not focused on AI chip companies, But if the traditional chip giants find opportunities, it will be easy to move into the field of artificial intelligence.
Just one year, the "Chip + algorithm" negative attitude of the prediction was Google overturned, Alphago achieved results began to stimulate people's nerves, while Tesla also began to layout AI chip, Huang that: ' Google's research on TPU and the results achieved through user learning proves that this is the right path, and that if you want your software to be better used, you should do your own hardware. '
At present, the horizon is designed and developed for its own application scenarios or smart camera applications in the chip, in 2017, the United Nations, some foreign manufacturers to the landing of the exploration, in 2018 began to try mass production.
2017 many chip manufacturers into the bureau, for AI start-ups, the most important is the screening requirements
At the end of 2016 to 2017, there appeared a large number of chip manufacturers, but many of them are just occasion marketing, hanging a name, ' some of the traditional chip manufacturers, he actually did not have in-depth thinking about this thing, whether the application level or the development trend of the algorithm. So he is more by this trend to the original things to the packaging, and then hang a name to do. ' said Huang.
And for another type of company, it did not make chips before, but it is more to the other people's GPU, or a DSP processor for a name, called their own artificial intelligence processor, but in essence the chip is not his own design, so all are in order to catch this upsurge, artificial intelligence plus chips.
Such a kind of self-deception of reform, in fact, there are many non-AI enterprises, from the current point of view, everyone is said to use artificial intelligence to other industries, and artificial intelligence appears to be like a ' snake balm ', and the most important thing for TOB start-ups is to identify the needs of your partner's value, The degree of rarity.
Huang that as a Tob artificial intelligence start-up company, every day you may see a lot of B-end customer needs, do not know what is the real demand, which is the false demand, while some users expect too much, even false demand, just to let you cooperate with it to do PR, so for a start-up enterprise, The most important thing is to ' identify the value of the needs of your customers, the degree of rarity, find the most suitable for you, and the most valuable place to focus on the dash, which is the biggest core problem. '
And in the field of artificial intelligence, the future unicorn certainly more than one, and to become a unicorn is the key to how the product landed, how to scale, Huang: ' For the current artificial intelligence in the field of start-up companies, when encountered problems, the previous experience and judgments in this changing world, In fact, many of them do not apply. Only continuous learning and observation, the future will certainly not only a so-called unicorn enterprise. The key to becoming a unicorn is the size of the industrialized landing that determines your value. '
The artificial intelligence is imperceptible to the life optimization, the prophase is not easy to perceive
The artificial intelligence is imperceptible to the life optimization, there are many not even easy to realize. When asked about the specific changes in the life of artificial intelligence, Huang told us his views on the actual landing of artificial intelligence, and his understanding.
Huang thinks, at present can let the user feel the real change has two: 1, low speed automatic driving; 2, Smart camera. Huang said, ' Low speed automatic driving will be a relatively fast and very deep (project). " Especially to raise him on a shared trip. It would be more straightforward to share a trip company to talk about it. ' At the same time, low speed automatic driving is the first step in the future, while the smart camera is the most easy to produce mass production, but also a gateway to the internet of things.
But other changes may be a subtle way to optimize life, a change that cannot be perceived but persists. Artificial intelligence will transform a variety of user experience, some of the world, some of the hidden world. ' The Life of the world, is to serve you, it can help you better to make the future decision planning, that is, it can provide businesses, to provide more hope for you to provide services for you to create convenience. '
At the same time, Huang that ' as long as the real economy does not arise, the attitude of capital to artificial intelligence enterprises is still good. ' Cause ' ai is known to be like this machine, the transformation of electricity and communications, his transformation of mankind will be profound enough to produce great value, you just watch this, you just agree with this, then I believe that the investment around it will never fall. '
Ai in Intelligent home is an interactive means, the transformation of smart home should be in the front
In the planning of the horizon, in addition to automatic driving and intelligent home, in Huang view, the whole problem of smart home is: ' deployment costs, and the next whole operation. '
Huang that the weak point of smart home is that AI provides only an interactive means, rather than a core value, the core value is content and services.
At the same time, the smart home does not currently have a common agreement, is essentially dividing the interests of, ' whether it is Haier or gree, or the United States, they want to monopolize this thing, do not need interconnection and interoperability. And this market is difficult to carry out a unified planning of things, there is no need. So this is really an objective fact. '
So for smart home, in the rear-mounted market deployment is actually very difficult, Huang that the most promising is in the front of the market, is in the renovation, or a new house when there will be opportunities. You will see more and more new houses will have more intelligent presence inside, light switches, including door opening, or air-conditioning control. '
Currently, the horizon on October 20 has completed the Intel-led billion-level a++ round of financing, but also in the active exploration of commercialization, Huang hope in 6, 7 years after the implementation of automatic driving. Recently, in 2018, 2019, some mass production was carried out. Tencent Technology
4. Skip depth Learning route associated memory AI Alternative
From Altera, Saffron, Nervana, Movidius to Mobileye, several companies bought by Intel seem to favour their ambitions to expand the AI map. However, how does Intel plan to integrate this? Especially the biggest mystery--saffron; The company uses an AI branching technique that is different from deep learning--associative memory AI. Perhaps from the interview with Saffron can grasp a little clue ...
Intel's $1 billion drive to promote artificial intelligence (AI) ecosystems is one of the topics that the processor's giants are talking about. Intel has accumulated a wide range of AI technologies through acquisitions and investments by Intel Capital in AI new ventures.
From Altera (2015), Saffron (2015), Nervana (2016), Movidius (2016) and Mobileye (2017), several companies that have been acquired so far seem to be in favor of Intel's ambitions to expand AI territory. Intel Capital also enriched its AI portfolio by investing in Mighty Ai, Data Robot, Lumiata, Cognitivescale, Aeye Inc., and Element AI.
However, it is still unclear how Intel intends to integrate all this. As AI innovation is still at an early stage, it should be reasonable to have a clear and decentralized approach to Intel's AI strategy. We may have to wait a while to see a more coherent development.
Intel discusses its AI hardware product portfolio more often than the overall AI strategy. For example, Intel has announced that it will ship the Nervana Neural Network Processor (NNP), formerly known as Lake Crest, by the end of this year. Naveen Rao, Nervana's former CEO and co-founder, described the NNP as a ' dedicated architecture for deep learning '. Naveen Rao is now the Intel Vice President and general manager of AI products.
In the AI chip, Intel also has other ' weapons ', including the Xeon series, acquisition of Altera acquired FPGA, car with Mobileye, and Movidius for the edge of the machine learning chip.
However, Intel has remained silent about the areas in which AI applications may actually focus on AI. Ai is, after all, a broad and in-depth technical field. The biggest mystery of Intel's continued expansion of its acquisition list is saffron.
Nearly two years after the acquisition of Saffron, Intel issued a "Intel Saffron anti-Money laundering advisor" (Intel Saffron Anti-Money Laundering Advisor) in October this year; AML), the news has aroused widespread concern. While the AML is clearly implemented on the Xeon processor, the product is not hardware, but a tool that helps investigators and analysts find financial crime.
"EE Times" has the opportunity to visit the Saffron Marvell Financial Industry Solutions director Elizabeth Shriver-procell, in-depth understanding of saffron products behind the AI technology, And what she sees as the benefits of Saffron as Intel Corporation.
Saffron Marvell Financial Industry Solutions director Elizabeth Shriver-procell on the other hand, what we want to know is that A long fight against financial crime, like Shriver-procell, is largely responsible for what is inside the world's largest CPU company.
Please talk about yourself first. I heard you are an expert in financial analysis, and have worked in departments such as the Ministry of Finance and many other companies.
Shriver-procell: I am a lawyer, mainly engaged in the fight against financial crime. I have worked in international consulting companies and major financial institutions. He left bank of America earlier this year and joined Saffron. Yes, I also worked as a project manager for analytical development at the US Treasury.
Did you use saffron products before joining saffron?
Shriver-procell: Some of the organizations I've contacted, including the clients of the consulting firm, have used saffron. I have always been interested in this platform, so when the opportunity arises, I firmly hold.
So what does Saffron offer?
Shriver-procell:saffron always think of the wide application of the customized ' analysis platform ' positioning for sales and publicity. Users include supply chains, banks and insurance companies.
What changes have been introduced to the Saffron platform after the ' Anti-Money Laundering Advisor ' tool was launched?
Shriver-procell: We are now launching more specific products for specific applications.
Associative memory AI
I guess the main reason Intel bought Saffron was to get AI technology, not just to crack down on financial crime (though it was worth it). Please talk about saffron design AI professional technology and applications, it and other AI technology is different?
The AI technology used by Shriver-procell:saffron is called the ' Associative Memory ' (associative Memory) method, which is a differentiated AI branching technique from deep learning (Deep Learning). Associative memory AI is very good at observing large and diverse data and identifying signatures or patterns from different databases that are far apart. It enables the integration of structured and unstructured data from enterprise systems, e-mail, networks, and other sources.
Take Mary, a bank client, for example. Mary travels to London every other week and goes shopping at the Liberty Street store. John, who lives in another country, went to London at about the same time as Mary, but his purpose seemed quite different. So what's the relationship between the two? What do you have in common? Can we view its IP address? Can you find anything similar in its login mode? Is there any indication that any unlawful action is in progress?
So, the point is that associative memory AI can view and complete so many seemingly unrelated repositories at the same time?
Shriver-procell: Not only that, it is also able to quickly accomplish a task that was extremely time-consuming. Although deep learning requires a lot of training, the associated memory AI does not require any training. This kind of AI branching technology can be learned quickly and without modeling.
Saffron's white box ai is mentioned in the press release, can you explain it in detail?
Shriver-procell: The so-called ' white box ai ', what we want to emphasize is transparency. It allows us to explain how to draw some kind of conclusion. In the past, financial institutions bought model-based fraud detection solutions to suppliers, which we call ' black boxes ' because users don't know how their software works in black boxes. And when regulators ask financial institutions to explain how to come to a conclusion, they cannot really explain. Because they can't see what's in the black box, they can't tell if it's working.
In highly regulated industries, the ability of financial institutions to provide information transparency is critical.
Sounds interesting, and seems to be quite the opposite of the deep learning AI. Some security experts worry that the car manufacturer could not explain why AI made the decision, such as turning, when deploying a deep learning AI in a self-driving car. The lack of transparency in the learning process makes it difficult for carmakers to verify the safety of self-driving cars.
Shriver-procell: I think the most important thing is to recognize that there are many different ways of AI. When the Intel CEO spoke about unlocking the vision and potential of AI, he suggested that we should try something new and explore new learning paradigms.
Do you think that the different AI branching technologies will eventually converge on the same point?
Shriver-procell: I think these branches are complementary. As we have seen the development trends of various fusion applications, I think that the combination of various types of AI to meet the needs of a variety of applications.
Please introduce more new products about saffron.
Shriver-procell: As previously said, Saffron always sells products on a platform. Now, as we identify specific needs in specific segments, we decide to start introducing specific solutions that can be challenged in different markets.
Saffron has been based on the experience of combating financial crime and occupies a very strong position in the financial market. By using the 360-degree view to complete structured and unstructured data, we are able to understand patterns found across data-storage boundaries.
We also announced that the Bank of New Zealand recently joined the Intel Saffron Early Import program. This is designed for organizations interested in providing innovative financial services with the latest advances in Lenovo memory AI.
What benefits do you think Saffron has gained after joining Intel?
Shriver-procell: The benefits of joining Intel are enormous. We are discussing the serious problems faced by large financial institutions. To provide strong support, the power and resources of a giant like Intel, as well as the full support of Intel as a technical partner, are needed to successfully build new features and applications on the Saffro platform and upgrade and extend it. With the rapid progress of AI, it is not to be neglected to explore new things and methods to realize AI.
Compilation: Susan
(Reference text: Intel/saffron AI plan sidesteps Deep Learning, by Junko Yoshida) Eettaiwan