Does AI really bring about the next change? If so, where is the opportunity?

What is the Guide to Artificial Intelligence (AI)?

· Will it really bring about the next technological change, what opportunities will it bring?

· Is deep neural network the ultimate way to achieve AI?

With this series of questions, the following article has been discussed, and the AI ​​generation has been compiled and compiled, and I hope that readers will gain something.

I. Introduction

· What exactly is AI? Why is it mentioned again widely by people?

· Does AI really bring about the next change? If so, where is the opportunity?

Is deep neural network the ultimate way to achieve AI?

· In what form will AI appear?

· How to participate in such a country, such a company?

Loading...

Second, AI / AGI is a form of organization of world knowledge

·Technical barriers create business value

·From retreat to pragmatic change

· AI is the core of the next generation search engine

· Opportunity = Technology Change / Upgrade & Scene Innovation

· Association is business value, finding connections in big data, artificial intelligence is the means

Third, integration + openness will become a technical barrier

·Customized

· Integration requires strength and imagination

·Exploration and follow

·Open guidelines

Fourth, the first step to build a moat: the concept of automated construction

· Integrated delivery process and technical architecture design

· Currently, AI can only be used as a tool

· In the future, it is best not to let AI form a closed loop

·Search and recall

Fifth, the second step of building a moat: rational planning of input and output

Sixth, the third step to build a moat: rethink the platform threshold and stickiness

· Threshold, sticky and open ability

·Build an AI platform

·pepper

Seventh, the fourth step to build a moat: all things connected and standardized sensors

Eight, the fifth step to build a moat: collaborative design and cooperation

· AI improves the overall efficiency of human collaboration

· The main obstacles for business users to achieve their business goals

·Big amount of data

·Working together on shared big data

· Coherent calculation mode

Nine, the sixth step to build a moat: to create a visual service

· The importance of visualization

· Carefully design visual interaction mode

Ten, summary

The tipping point where artificial intelligence is once again heatedly discussed seems to be autonomous driving, which is the author's personal experience. Major mainstream automakers have given their time to put their respective products into use, and even many Internet companies that are not familiar with the auto industry are actively participating.

Automated driving and advanced assisted driving technology are the fusion points of multiple research directions, and there is more than one technical path. This time, people want to let the car have artificial intelligence through deep learning. This end-to-end ideal state seems to make the problem simple, but because scientists still can not fully grasp the specific situation inside the black box, it is even more difficult to know the similarities and differences between the black box and the human brain, making this thing uncontrollable.

At this point, the author feels a lot of problems. The study must be rigorous, and it is necessary to have a correct methodology and product view. But fundamentally, maybe for AI, there are several issues that need to be rethought:

What is AI? Why is it mentioned again widely by people?

Does AI really bring about the next change? If so, where is the opportunity?

Is deep neural network the ultimate way to achieve AI?

What form will AI appear?

How to participate in such a country, such a company?

View

· AI / AGI is a form of organization of world knowledge

Technical barriers create business value

The most famous description of artificial intelligence is the Turing test:

A person does a series of questions and answers with the other party in a special way without touching the other person. If, for a long time, he can't judge whether the other person is a person or a computer based on these questions, then the computer can be considered intelligent.

Artificial intelligence can be applied to many fields, such as AI + chip, AI + sharing economy, AI + optical projection, AI + knowledge data + speech technology / + emotion calculation and so on.

From retreat to pragmatic change

More important than the abstract concepts above, how to participate effectively and create real business value. For example, to develop a self-driving car, the author focused on how to do it from the beginning, to think about why, but now the idea is, "I don't need a car that I can run by myself. What I need is in the process, gradually Create your own moat and create business value."

The author believes that in this field, although the time of entry is first and foremost, there is no such thing as a permanent follower or a corner overtaking. In a market where you can see the top, it is easier to be a follower than a explorer (such as Huawei and Ericsson in the communication solution industry); but in an unknown field, the meaning of being a follower is not big, everyone, every company All should have his own AI view, looking for those quick opening and closing windows, and building your own core barriers is the key.

AI is the core of the next generation search engine

When it comes to search engines, you have to mention Google and Baidu.

Baidu's unmanned car has already been on the road, and its implementation is far from artificial intelligence. However, the Automated Driving Division was established in December last year, and the L3 Business Unit was established in September this year. The two are mutually reinforcing and intend to apply AI to autonomous vehicles and ADAS products. The reason why Baidu invested so much is probably because AI can be said to be the core of the next generation of search engines, which will be elaborated later.

Loading...

Baidu unmanned vehicle

Opportunity = Technology Change / Upgrade & Scene Innovation

Opportunities arise in two situations, one is the natural change and upgrade of technology, and the other is the innovation of usage scenarios. For example, the search engine of the PC Internet era, such as the smart phone in the era of mobile Internet.

The upgrade of a search algorithm has established Google's dominance for more than a decade or two; a scene innovation with touch-based interaction and upstream and downstream integration allowed Apple to capture most of the profits of the smart phone market, even for the Yangtze River Delta, etc. The region has brought many industry dividends.

Association is business value, finding connections in big data, artificial intelligence is the means

Mobile Internet has accelerated the arrival of the era of big data, analog signals seem to disappear from people's lives, and more and more activities are expressed in digital form. The amount of data has skyrocketed; speed, variety, and uncertainty are also growing; most data, such as images, video, natural language, and symbols, are unformatted.

If there is a computational model that can process and understand them, and find a connection between these disorganized data, companies/entrepreneurs can improve production efficiency, improve business operations, and even perceive on the basis of this computational model. Innovative scenarios in forecasting, reasoning, or thinking; perhaps the opportunity comes from here.

Intuitively, companies that can own and provide this kind of computing model have a certain amount of data and a well-designed open platform solution. Below, the author will discuss this in detail.

The amount of data is important but not the bottleneck. The bottleneck is the ability to integrate/integrate. Baidu and Alibaba have done the same thing. Baidu's ADAS and self-driving cars and Alibaba Cloud will generate several T data per unit time, including user behavior statistics in various aspects such as voice and operation. This data is reported to The cloud will become the most important data channel for understanding users and training networks.

Now consider such a question: In the field of artificial intelligence, what are the advantages and disadvantages of small companies compared to large companies? The author believes that small companies have no advantage in building an AI platform and standardizing specifications. However, once the platform is built, small companies have advantages based on the scenario innovation of this platform, because even if the company is small, it is creative. No size.

Here, the author divides the size of the company by the amount of data. Large companies have more user data, but the larger the amount of data, the less marginal benefits. Therefore, the startup only needs to obtain enough data resources. As for the difference in data size, it will not bring about a substantial gap. Based on this, the author believes that the amount of data is not the most critical. The key is creativity and Computing resources.

· Fusion + Openness will become a technical barrier

Custom

If there is such an architecture, companies/entrepreneurs are free to choose their input combinations, such as sensor data, images, video, natural language, etc., through the processing of one or several computational models, output the knowledge of the standard format. . Enterprises/entrepreneurs therefore need to consider how to understand the inherent structure and connections of unknown signals, and directly enter innovations in application scenarios. This ability to integrate + open will naturally become a technical barrier.

It is important and difficult to build multi-source integration capabilities. How difficult is it? In the past, the author introduced in detail the method of multi-source sensor data fusion in the article of automatic driving.

Integration requires strength and imagination

At a higher level, there is another key word for integration, which requires imagination and strength:

Apple uses a variety of sensors, such as capacitive touch screens, LCDs, cameras, and gyroscopes, to create a usage scenario where touch is the primary interaction. None of these hardware is produced by itself, but its engineers can communicate and guide in depth with various hardware vendors. When I got the iPhone 4 for the first time, my heart was shocking. I felt it was a work of art, which subverted my previous understanding of the mobile phone and made it so natural and comfortable to slide. (At that time, there was only one department manager in the whole department. He gave me the phone very generously. "You are very geek, so now you can help me crack this phone, grab the log, I need to know why. Ok, then I turned around without even giving me the data line. Of course, this is now a joke.) At that time, I realized that a good product is the same hardware, even if you use the same vendor. The group can't achieve its effect. This kind of integration strength not only shocks the users, but also makes the competitors do not know how to start. Behind this is the imagination and strength that the author mentioned.

Creation and imitation

In the AI ​​era, both the explorer and the followers will become more difficult to create and imitate, especially followers: now, you can also disassemble the competing hardware, analyze the software interface released by the opponent, and even use the high-speed camera to reverse The inference algorithm, in the future, is faced with a black box, the input is the data from which I do not know, the output is the data after hundreds of millions of nonlinear operations, the followers will have to completely forget the reverse engineering (I mean Is a series of technical/non-technical means).

Open guidelines

In the work of designing SDK/API, the author summarizes and adheres to a criterion: N is provided externally, to be restrained, to control the interface granularity, and let the user create the effect of 1+1 》 2 . This is a test of the design capabilities and vision of the R&D staff, and it also demonstrates the openness of a platform.

Integration and openness seem to be fighting with each other, but in reality they are in each other. To be such a person: there is an architectural view of integration and integration, and a vision of openness and restraint.

The author believes that: good products = core technology + integrated delivery solutions, these two aspects will form a "moat" that competitors can not easily imitate. Below, the author will analyze in detail how to do both.

· The first step in building a moat: an automated view of automation

Integrated delivery process and technical architecture design

How to participate deeply in it and realize personal value? In addition to the above-mentioned view of architecture, there should be the ability to communicate honestly and discover needs. This is what the author has summarized from his early work. When asked: "What can I do for you?", you can always find real needs. The integrated delivery process and technical architecture should be designed based on real needs.

For example, when a computing module is provided externally, some people can implement and package the functions that can be conceived into an interface to provide users with the functions and subsequent technical support problems in the complicated technical documents. I didn't think about it, and even some functions depend on the user to actually experience it in actual use. It seems that I have done everything for the user, but it adds unnecessary trouble. This is an explanation of the word restraint that I mentioned above.

The author solves this problem by communicating with customers, finding demand points, redesigning the delivery process and software architecture. In the communication with the user, the author found that: 1 customers need highly customized services; 2 customers will choose the competing products because of the size of the delivery software package (early, limited by the storage capacity of the smart machine, the software size is Issues that must be considered.).

Therefore, the author designed the overall delivery mode: the user only needs to make a choice on one page (some users will choose function A, function B, while others will choose function B, function C, etc.), when the user completes the customization. After that, the automatic build tool will grab the corresponding module code in the background, dynamically package the package, and deliver it to the user. The software thus delivered, users of different needs get a highly customized package. The entire process is fully automated, and no one is involved, ensuring product quality.

In the AI ​​solution I envisioned, there is a similar design in the scenario of external integration to deliver AI computing power: At that time, the automatic build tool captures the corresponding computing model, and the entrepreneur can pass the author based on this. The simple custom interface mentioned is connected to different computing models, which is even more challenging!

The author believes that in the AI ​​era, high automation is the first thing that should be considered. In the past, to achieve this goal, we must fully consider various conflicts and coupling problems when designing the first interface and writing the first line of code. In the AI ​​era, these problems have been magnified or changed, and even the experience accumulated in the development of systems in the past has become no longer effective: how to deal with the fusion between computing models? How to deal with border issues? These are issues that industry and academia need to solve.

Currently, AI can only be used as a tool;

In the future, it is best not to let AI form a closed loop

The intimacy of the neighborhood is a very interesting psychological activity. With the artificial intelligence being mentioned again and again, with the innovation of technical means, the outline of artificial intelligence seems to be more and more clear. I sometimes think that once one day it breaks through the storage limit and learns to move between networks, then It’s terrible. As stated in the overview, deep neural networks make this thing even more uncontrollable.

If artificial intelligence is the goal, deep learning is now a means, a computational model mentioned above. This means of implementation has not yet been finalized. It can even be said that it will change drastically in 2017. This trend will be discussed in detail in Chapter 2.

Therefore, the author wants such an artificial intelligence, which is both a powerful tool and a form of knowledge organization. It automatically correlates seemingly unrelated signals. It subverts the past data storage form. It is before me. The Internet brain mentioned in the article, the brain of 2017, whether it is a deep neural network or a tree or something, will usher in dramatic changes. If it goes well, maybe it will form a memory and create new knowledge. If one day it evolves into the organization of world knowledge, it is the core of the next generation search engine, and AGI is not a dream.

Search and recall

Then, let's make bold assumptions. The Lenovo brain is always active in different regions when dealing with different tasks. This kind of world knowledge organization system should be composed of countless small models. The small models are highly connected by software technology or hardware technology such as weight sharing. The models are weakly connected and form a complex topology. At this point, the automatic build tool described above by the author evolved into a high-performance block search tool, a search engine. Quickly capture computing models, dynamically isolate and package them to users. It's like a person remembering a skill he has mastered.

· The second step in building a moat: rational planning of input and output

The input and output of a system directly affects its important properties such as ease of use and subsequent scalability. This problem requires specific analysis of specific problems.

· The third step in building a moat: rethinking the platform threshold and stickiness

Threshold, sticky and open ability

I have been engaged in software research and development since 2008. I have experienced several platforms. One experience is that the more the platform is low, the more developers can be attracted, the more developers, the platform is naturally prosperous, this is an Au Pair. Mutual benefit mode. It seems to be simple. In fact, it requires a detailed technical route and a complete set of solutions.

For example, Google's Android platform. When the smart machine emerged, the author quickly switched from Qt to Android, and even the programming language switched from C++ to Java. During this period, it did not experience too much pain. This platform turned the app development from a few people's games into a group of people. Now that Android has passed its rapid growth period, the author has a lot of disappointment when leaving the platform, such as the language and tools used; the test machine is easy to obtain; the system code is open source, and the common growth of these years, leaving is Since the martial arts and so on. If you don't think carefully, think that the operating system is too heavy in the AI ​​era, you can't easily leave.

-- These are issues that are worth considering and drawing into when building an AI platform.

Pepper

Pepper robot

In August of this year, Alibaba joined Softbank Foxconn to form a robot company, and Aliyun also formed its own NLP team. They quickly ported yunos to this robot called Pepper. At the same time, the various capabilities of the robot are packaged into SDKs for developers, and the processing between various capabilities is transparent to developers. This is also a form of fusion + openness.

This 120cm robot is easier to walk into everyday life and seems to generate business value faster than autonomous driving.

If this robot is considered as an integral part of Alibaba's artificial intelligence strategy, once the sensor-rich hardware has the OTA capability, it will become an AI ecological environment, which seems to take the lead.

But the author also has another opinion:

First of all, this technical architecture is not fast-portable;

Second, in the era of AI, the era of light OS, OS can no longer form technical barriers.

A single hardware frame defines the scope of application, and limiting innovation is only a matter of time.

To provide SDKs based on large hardware platforms or specific hardware platforms, it is necessary to consider many issues such as open security, reasonable efficiency, and scheduling consistency. It is equivalent to re-developing os.

Ali envisioned a robotic ecosystem and designed a complete solution for this, and developers are innovating on this hardware. The author believes that a specific form of ecology should not be created in order to create ecology. The ecology should be formed automatically and evolved on its own.

The author wants such an AI environment, entrepreneurs use its open data fusion capabilities, broad innovation, seemingly no ecological environment, but it is a real technical barrier. Another benefit of this is that the technical path is quite clear and can have its own rhythm. The author believes that technical barriers and technical paths are quite important.

· The fourth step in building a moat: the Internet of Everything and standardized sensors to create a sensory brain

When designing an AI platform solution, you should always keep in mind AI +, because AI can't form an industry by itself. It needs data education in various industries. Once this platform is built, AI will evolve into technical barriers in various industries. This barrier adds value to the industry.

The AI ​​+ smart home, finance and medical industries are the closest to everyday life, and the data quality of various sensors directly affects the AI ​​platform. It is necessary to plan various sensor data fusions in advance and to standardize sensor data standards as well as communication standards.

The perceptual brain is the stage that artificial intelligence is bound to experience. After the data is merged, a higher concept is abstracted. This brain has cognitive ability, and the next stage is the cognitive brain.

The fifth step in building a moat: collaborative design and collaboration & creating visualization services

Artificial intelligence can improve the overall efficiency of human collaboration and enhance human cognitive ability:

"Like the railway and wireless communication technologies, artificial intelligence will completely subvert people's existing behaviors and improve the overall efficiency of human collaboration. At present, no science and technology can affect the huge amount like artificial intelligence, even if it is claimed to be Million times of quantum computing that increases the speed of calculation. Because artificial intelligence enhances the reachability, it changes the way of connection between things, and this cannot be simply quantified."

The efficiency of collaborative work across corporate data across shared data is a major factor affecting business objectives

Big amount of data

When I was responsible for designing the technical architecture of a product user behavior data analysis system, the author focused on the scalability, robustness and reliability of the client. This system is very efficient and accurate in the early stage of the product, but with the increase in the number of users and the number of reported data, the author found that the background personnel spent considerable time on data cleaning.

Earlier, Baidu’s unmanned car ran a day’s data, requiring hundreds of servers in Baidu’s data center to process it in a week.

This is a challenge that the big data era will face.

Work together on shared big data

Even more troubling is that for developers and enterprise users on the AI ​​open platform, their data workers are independent of each other, use different programming languages, have no common concerns, and waste time on data collection and cleaning. on. So how to collaborate efficiently on shared data will be a major obstacle for developers/enterprises to achieve their business goals.

Coherence calculation

As a technical person, collaborative design used to be a direction that the author was obsessed with. Tencent is a very open company that allows employees to take their work home, so think about such a usage scenario:

1. I don't believe in the security of network transmission (although the enterprise cloud disk is very convenient) - network isolation

2. I also don't want to manually put the content that needs to be done into the U disk one by one - physical isolation

3. Even the computer at home doesn't want to keep it connected at all times - security guarantee

4. I want a safe and efficient fully automated tool - high automation

5. Can take snapshots of current work in a very short time - high time sensitivity

This way, even if there is no network connection in the home computer, the screen will display the work being done, the cursor is staying on the line of code/document/graphic that was being written before, and the real seamless switching. In the end, the author completed the development of this software based on the idea that everything is data. This happened a few years ago. At that time, I was arrogant that personal idealism and commercial interests could not coexist. In the end, the software was only used by myself and several small partners, which doubled the time utilization. Of course, from the approval of winning small partners, the author also realized personal value.

The issue of synergy in the design of the AI ​​platform should be fully considered. Any problem of time utilization should not be ignored.

Think about this seamless switching problem again. With more developers on the AI ​​platform, our computing power will cover a considerable proportion of consumers. If consumers move from indoors to outdoors, moving from one scene to another, there is a consistency of experience behind it. Problem - Different entrepreneurs cite our different computational models. There should be a computational model between models that can run through time and space, so that there is no comparison between multiple computing models between consumers, consumers and computing models. Slot switching, can this calculation mode be called: coherence calculation? Or immersive calculation?

When it comes to coherence calculations and scene switching - distributed computing is also coming out, is there a feeling of a new bottle of old wine?

· The sixth step in building a moat: creating a visual service

Carefully design visual interaction mode

Man is a kind of visual animal, expecting feedback from all things. Think of AI as a tool, and it should be a reasonable feedback and fault tolerance mechanism for it.

Perceived causality affects each other between the five senses, and what they hear will affect the sense of touch. The longest time interval between consecutive events that feel that one event produces another event is 140ms. This time interval is the longest time limit for perceived causation. If your product reacts to a user's action for more than 140ms, then the user will come out of an unconscious and efficient state and think about whether the reaction is caused by his own actions. If the user taps the character after more than 140 ms to display it during the use of the search box, the user will feel that the word is not input by him, and his attention will shift from the meaning of the text to the action of the character input, resulting in The speed is reduced, and the action that can be handled automatically by typing becomes an active consciousness process, increasing the probability of user error.

The above paragraph is taken from the project summary report of the author's touch experience problem of intelligent machine products. The author has developed the technical path of improving the touch experience by optimizing the LCD display effect by studying the characteristics of brain constants in cognitive psychology. Complete project public relations. The core idea is: to solve a specific problem, if the most direct path will be limited by the current technical capabilities, it is also possible to use the trick.

EIW/A Enameled Aluminium Wire

About EIW/A Enameled Aluminium Wire

SWG enameled aluminum round wire for household appliance heat sink, Polyimide enameled round aluminium wire, class 220. Package: 250*500/250*400/250*600 wooden spool.

EIW/A Enameled Aluminium Wire


Product Name

Enameled aluminium wire EIW/A 1.20-3.00,CLASS C-220

Material

Aluminium

Type

EIW/A

Themal class

calss C -220

Country of origin

China(Mainland)

Used

motor,transformer,rectifier and such




Aluminum Winding Wire,Eiw/A Enameled Aluminium Wire,Reasonable Price Enameled Aluminum Wire,Enameled Winding Wire Pew

HENAN HUAYANG ELECTRICAL TECHNOLOGY GROUP CO.,LTD , https://www.huaonwire.com