OPPO RENO7 Pro Review: Super Light Cat Eye Shooting

After upgrading more products, the RENO, which has changed from the R series, has also found its most good way: color value is supreme.

To start it, I look at it, look good (excellent design), it looks like I am fine (selfie).

This time, the latest OPPO Reno7 Pro revisited these two aesthetic pursuits, made a new upgrade. Next, let’s start from it.

Although the design of mobile phone is homogeneous, OPPO Reno has a unique and bright, which allows us to recognize the Reno model in the sea.

The design guides can be said to be the routine operation of OPPO Reno. In the Reno6 series, in order to achieve a sparkling "summer sunny" effect, the emergence of the drill 3.0 process is promoted.

The design inspiration of the Reno7 series owner’s color "Star Wish" should be from the natural scene from "Meteor Rain."

In order to "leave" on the back of the fuselage, OPPO chooses the secondary lithography on the basis of the crystal diamond process, introduces the industry’s first LDI star rain photolithography process, leaving 1.2 million on the glass backplane. The road width is not a fine raster line.

These fine raster lines created a grand-explosive meteor rain landscape, making a meat meteor shower, tentacles available.

With the double-layer double-plating process, the back of the Reno7 Pro can see that the cold and alternating gradient color is like a meteor shower, and the sky is still changing in the sky.

In addition, the grating texture created by Star Rain Lighting Process also brings a unique delicate and frosted touch, which can resist the erosion of fingerprints and oil, bringing long-lasting and comfortable grip.

▲ OPPO Find x3 Pro microscope under the back of the OPPO RENO7 PRO

OPPO Reno7 Pro Thermal steering uses the right angle in the right angle to add the design of the near-flat glass panel in the opposite side, and the view is more angularly distinct, and the bifunction of the surface of the matte surface is used in the straight side.

Although the Reno7 Pro uses the angularly distinctive straight scholar design, OPPO has added a CNC fine cutting step in the upper and lower sides of the box in the aluminum alloy, and the 2.5D micro-arc transition design of the front and anti-two-sided glass is used. The appearance is still available to provide a relatively rounded edge transition.

Moreover, the straight son of the grinding texture also enhances the stability of the grip, no hand, no hand, but it is very known.

Although the straight-edge design will make Reno7 Pro think more than the previous generation, the thickness of 7.45mm is actually a slightly thinner than Reno6 Pro’s 7.6mm thickness, plus Reno7 Pro machine about 180g Weight, the feeling of holding in the hand can be described in the "light" two words.

After the rear image module part, OPPO Reno7 Pro first adopted a laminated design of "Gull Integral", that is, the matte layer is suspended from the back of the fuselage layer, and there is also a layer of smooth and bright surface overlap together. . Moreover, there is also a ceramic lens wrapped in the matte layer to lift the lens scratch resistance.

As a "God of God" on the design of the whole machine, OPPO hides a unique "garage breathing lamp" in the 1mm polarized space of the rear imaging module.

Thanks to the "fiber hot technology" in the industry, OPPO can not only use this extremely narrow 1mm space, but the star cyclic breathing lamp has a discontinuation of the end of the back module. The lamp can be evenly lighted, and it is soft like a star ring.

The application scenarios of the breathing lamp mainly include information notifications (slowly flickering breathing lamps), and the caller reminder (fast flickering light effect), the state of charge (often lit). The user can set the "Star Ring Respi Light" in the system settings, such as the application that needs to be notified, and the time period illuminating the breathing lamp.

Give it a charge, look at this flashlight, it is in my eyes, it will really shine.

Amazing place, as well as the 6.55-inch OLED flexible screen on the front of the OPPO Reno7 Pro.

In addition to supporting the 90Hz screen refresh rate, its four-week border uses extremely narrow design, raising the screen to 92.8%, creating the effect of "positive full screen", bringing stronger view.

And on the basis of this screen size, further narrow the body area, slightly reduced the difficulty of single-handed control.

The straight-edge screen design is a very narrow border, which can be visually able to bring very shocking and immersive visual effects, which can be said to be the gospel of "straight screen enthusiasts".

Then, let’s take a look at us in Oppo Reno7 Pro, how good is it.

OPPO is again combined with Sony, and the depth is customized with a 32 million-pixel sensor IMX709, and the starters are loaded in the front lens of the OPPO Reno7 Pro. OPPO will "permanently exclusive" custom sensor, and call it "pre-supermodight cat eye lens."

The reason for the "Super Light Cat Eye" is naturally a photosensitive characteristic of the cat eye. As we all know, there is a similar mirror’s "clear blanket" reflective structure in front of the mirror. This is that the cat can only be seen in the six-third light volume of the cat. You can still have it in the dark light environment. Clear vision.

▲ OPPO RENO7 PRO shooting

The Reno7 Pro’s pre-cat eyeliner is joined the RGBW arrangement and algorithm of the self-developed, the purpose is to significantly lift the amount of light of the lens and enhance the imaging performance under the dark field. Simply, that is, dark light will be brighter, and less noise will be less.

OPPO and dark field scenes are more hard, it is no longer a good thing. Missing OPPO has been equipped on the OPPO R7 PLUS in 2015, it has been equipped with a process and core power of the process and core force that is still not mature at the time, and has not been further commercially available. However, obviously, OPPO did not give up the development direction of this technology.

On the OPPO Imaging Day in August this year, OPPO demonstrated the "RGBW Fire Sensor" technology based on 22nm process technology. This technology is now presented by the OPPO Reno7 series.

OPPO’s new generation of self-developed RGBW four-in-pixel fusion algorithm is introduced into 2 W pixels in each group, so that each pixel can have R (red), G (green), B (blue) And the recognition ability of the W (Mingdu) signal.

Since the W pixel does not bring any color, only the light sensation is only, the inclined amount is increased by 60%, and it is also reduced by 35% noise. Under dark light shooting scene, Reno7 Pro can make the screen more pure.

▲ The outdoor environment of the night, the face is weak, especially for mobile phone front lenses

In order to improve the efficiency of the algorithm, reduce hardware power consumption, OPPO to reach a deep partnership with Sony has written a series of algorithms such as self-developed quadratic pixel polymerization to the hardware of IMX709, forming an "image processing unit".

Not only let Reno7 Pro’s front cameras can preview imaging results in real time, but also automatically call the moonlight night view algorithm in dark light, and automatically call the Live HDR algorithm in reverse light.

More importantly, the algorithm is cured into the sensor, and the third party application can better call this sensor’s ability when calling the camera module. There are already a variety of applications (including WeChat and Zoom). It can effectively improve video call quality.

In the female child, the front camera viewfinder of the OPPO Reno7 Pro exhibited a high brightness preview screen, which is clearly visible. In the supersight cat eye shot, you only need to manage the expression in the screen, and then you can press the shutter button in confidence.

▲ OPPO RENO7 Pro front camera shooting, default auto mode / AI scene open / flash off

In addition, it may be because of the joining of the W pixel point, but the face makeup can not only be clearly bright and bright, but the color of the portrait has become white and very natural, bright.

▲ OPPO RENO7 Pro front camera shooting, portrait mode / AI scene open / portrait spot filter / flash off

Of course, in the dark environment, the famous RENO "Portrait Spot" filter can also be used smoothly. The new generation technology brings a sequential composition of the AI ??hair, which can be finely highlighted, and even the originth of the filament is clear. It can be seen, but also build a depth spot spot effect under large aperture lenses, and it is full of atmosphere.

If you really think that the ambient light is extremely dim, we can open the "flash" feature of the front camera, the screen will adjust to the highest brightness, and lit a white background to act as a fill light, illuminate your face.

In addition, the new "nude makeup" beauty algorithm on the OPPO Reno6 series, and the Ai Renewality 2.0 and other technologies have also been performed perfectly by the reno7 pro.

Clear nature should be the first view of the OPPO Reno7 Pro portrait photo and the video gives people.

Reno7 Pro built-in beauty algorgage can accurately identify the user’s facial position, and in the case of retaining the skin texture, accurately eliminate the possibility of the face, there is absolutely will not mistake the beauty to make acne cleaning, and finally reflect Out of a natural aesthetic effect.

It is especially obvious that the progress of the OPPO Reno7 PRO-like light spot filter is very large. The distinction between the main body and the background is more accurate compared to the previous generation, and the effect is more natural.

The "Emotional Filter" function is also provided in portrait mode, and the application of people ‘s split algorithm optimizes the skin color, while increasing the background of the background light source, and cutting personalized emotional expression.

The "Beauty" function in the Reno7 Pro portrait is a special characteristic that the little sister experiences after the experience. The rich makeup options provided by the portrait camera can automatically put a piece of makeup "stickers" on your face, let you automatically paint the eyeliner, eye shadow, lipstick, nose, blush, and even automatic Stick the sequins near the eyes.

▲ OPPO RENO7 PRO’s beauty imaging effect, received a high praise of the little sister

With this "AI Beauty" function, the little fairy woke up in the morning, and when it is placed on his face, it can take a self-contained "Good Morning Worker". Energy friends circle Selfie or video.

The beauty algorithm mentioned above can basically be applied to portrait video shooting. Portrait video section, Reno7 Pro has been optimized for night scenery, backlight, blur, spot, beauty and other aspects, helping users to shoot textured video.

It should be noted that the beauty and RGBW pixel four-in-one algorithms are optimized around the front camera, while the post camera is slightly inferior to face optimization, some portrait functions are also exclusive by the front camera.

After the rear camera part, OPPO Reno7 Pro is equipped with a 50-megapixel IMX766 CMOS as a wide-angle main photo, and there is an 8 megapixel super wide-angle lens and a 2 million pixel macro lens.

This IMX766 is also an OPPO and Sony co-customized sensor, which has 1 / 1.56-inch photosensitive area, F / 1.8 aperture, which has been loaded by a variety of OPPO flagship models, is a long-term tuning and consumer market. Flagship sensor tested.

It is worth mentioning that the spotlight of the main photo is more than the previous generation, and the same position can accommodate a broader landscape.

▲ is taken by the OPPO Reno7 Pro rear mirror group, and some pictures are compressed due to site upload restrictions.

From the perspective, the performance of Reno7 Pro can be satisfied, and the color performance is maintained in an excellent level. Whether it is a sufficient scene of the daylight, it is a scene of night darklight, and the performance of the back main photo is large. use. The AI ??scenario enhancement algorithm tends to see the eyeballs to see some scenes according to some of the scenes, and will increase the color saturation of some pictures.

Only the lack of telephoto lenses, the main burden of imaging is placed on the main photo, and the zoom in the scenery needs to be cut on the basis of 50 million pixel main photos. When shooting the vision, the analysis of the parsing and texture will be slightly It is insufficient.

Finally, let’s take a look at the configuration section. OPPO Reno7 Pro is equipped with the 6nm process process, which supports UFS3.1 flash, up to 12 + 256GB configuration versions.

The fluency naturally used in everyday use, facing some mainstream large mobile games, such as the "King Glory" and "League of Legends" in the MOBA class, Reno7 Pro can achieve "full frame" status; "Original God" is relatively high, and there is also a smooth operation. The average frame rate is about 50fps. Only some of the frames occasionally on the relatively complex picture scenario.

Reno7 Pro supports the "Extreme Steady Frame" function, allows the phone to automatically transfer system performance management according to real-time frame rate, temperature, etc. to ensure that the front game can run smoothly. Plus the VC homogeneous plate inside the body, its temperature control is not bad, after the high pressure is running, the body temperature is only a warm heat.

It is worth mentioning that OPPO Reno7 Pro has passed the test certification of the League of Legends Mobile Tour Event Machines, becoming a 2021 League of Legends mobile game event.

Within the body of a thin 7.45 mm, a large battery with 4500 mAh is equipped. At the same time, support 65W VOOC super flashing, 20 minutes can charge the battery to 80%, and can be filled with the battery about 30 minutes.

Keep the value is supreme, and it is slightly smooth, which is the impression of this version of OPPO Reno7 Pro. For me, light and staminated straight sides of the box design plus a nice narrow edge box straight screen, directly in the sense of touch.

On the other hand, in addition to the upgrade of the design, the startup of the custom RGBW front cat eye super sensation lens, is also the brightest place in Reno7 Pro.

From Reno 5 AI Video Beauty to Reno6 AI Video Makeup and Spotstown, and now the Reno7 series is now enhanced by self-study algorithm and custom RGBW sensors, OPPO Reno has always running on portrait video tracks.# Opporeno7 series #

China Unicom: Hit the core security guarantee combination, for the winter Olympique network gathering

How to play the advantages of the central security operations in the core network, protect the core security of Beijing 2022 Winter Olympics Zhangjiakou District? The online security portfolio is the answer given by the China Unicom North Region Operation Center (hereinafter referred to as "Northern Region Operation Center").

"The North Region Operation Center is the first major area operation center in China Unicom network. It is responsible for the core network maintenance of the five provinces, including Hebei, so you must shoulder the safety guarantee of Zhangjiakou 5G core network and VIMS core network. Work. "Wang Xin, general manager of Yunnan Operations Center, China Unicom Group Company.

Plan in advance to ensure that the network is in place

During the Winter Orance, the live floating has high flow density, large network demand, and is affected by new crown pneumonia, many people will watch the game through the network, so we must ensure that the network capacity can meet the needs of business growth.

"In order to prevent the network fluctuation caused by cutting-expand to affect the stability of the core network, we accurately investigate users’ needs, consider the increasing amount of more than 5G users in Hebei, and also take into account the shortcoming of users during large events."Zhang Yu, general manager of the North Region Operation Center, combined with the existing hardware foundation, and they put forward reasonable expansion planning, and actively communicate with the Group’s Winter Olympics, Hebei Unicom, Zhongxunyuan, equipment manufacturers, etc. Expand planning implementation.

At the end of October, the North Region Operation Center effectively reduced the occupancy rate of the core network elements of Hebei Province by pulling up a multi-set network element equipment, achieving the large-scale growth of NEM, SMF and other network elements, and effective reduced the occupancy rate of the core network elements of Hebei Province. Let the network have enough surplus space. Carrying the amount of traffic during the Winter Olympics, guarantee the online moment.

Willing safety line, building safety Winter Olympics

"Dedicated to the world, a simple and highly wonderful Olympic event, Unicom is unable to spend." Wei Haibo, general manager of Shandong Unicom, said, under the guidance of the group company, Shandong Unicom established the Winter Olympian network security leading group and working group, through formulating specific programs, Clear mission division, refine work measures, and ensure that the Winter Olympics is not lost.

A reasonable talent array can make a work less than half a effort, so it is necessary to change from the thinking shift from "doing things" to "employ people".

Zhang Yong, General Manager of Shandong Unicom Network, revealed that the core network security work in Zhangjiakou Fair is taken over by the network, and other relevant departments (including the North Region Operation Center) closely cooperates, through the common combing risk hidden dangers and combines the actual situation of Winter Olympics. To develop a rectification plan.

It is reported that this rectification work has been completed at the end of November.

In addition, understand the overall context of network security in Beijing Winter Olympics, and the development of follow-up work is significant. To this end, the North Regional Operation Center sent a talent to participate in the participation of China Unicom Beijing 2022 Winter Olympics and Winter Disabled Olympics. Develop a whole process through a small participation program, accumulating work experience for the Winter Olympics.

Innovative security initiatives are a must-win magic weapon to win the Winter Olympics.

At the safety mean, the North Region Operation Center not only completed the risk assessment and hidden dangers of Hebei Unicom network assets, but also in combination with Hebei Unicom to carry out network security emergency drills, including 5GC UPF / UDM, centralized VIMS and other networks in Zhangjiakou.

It is understood that 5GC UPF is a 5G core network user network element. UDM is the core network storage user data network, centralized VIMS is a network that deploys the Volte voice service in the center of the region. Guaranteed these nets have no loss of Winter Olympics, so it is not easy to carry out safety emergency drills in time.

The relevant person in charge of the 5GC team of the North Region Operation Center said that since the 5G core network adopts the SBA (service structure), NFV (network function virtualization), SDN (software definition network) and other new network architecture and technology, There are very few differences from traditional core network equipment.

Practice real knowledge, face new situation, new challenges, they dare to technically innovation, have completed millions of grades, thousands of-level cross-DC disaster recovery drills, will improve the safety of the safety exercises as guaranteeing the safety of Winter Olympics Important means.

The core network is the central part of communication network transmission, and its security is critical to 5G network security and smooth operation. By using standardized solutions, the North Region Operation Center built the "safe Great Great Wall" of network protection.

Collecting the soldier, open the 5G guarantee link

Whether it can be completed outstanding, the key is in the person.

Zhang Yu, general manager of the North Region Operation Center, revealed that in order to build a fully efficient Winter Olympics service support team, they start from the actual appeal of Winter Olympics, using the actual appeal of the core network operation and maintenance personnel, using the large-scale, efficient service Support mechanism, assembled a number of professional soldiers and other professional soldiers such as people’s network, Internet, signaling network, and voice network.

"Our team has established a full-scale communication and exchange mechanism from the Group’s Winter Olympics to the land and maintenance personnel, and the working mechanism has guaranteed." Xie Shaofu, deputy general manager of Dongao Service Support team, said Xie Shaofeng, deputy general manager of North Region Operations, pointed out that The emergency alarm tips of the local core network have proposed to respond within 5 minutes, and the fast security emergency treatment plan for the solution is given within 1 hour.

In addition, the support team uses the intelligent data monitoring large screen developed by the North Region Operation Center. By breaking the data isolation of different types of network elements, different manufacturers network elements, it is convenient for the network data. Metaire analysis, while using automation wind control tools to achieve a rapid response to the abnormal abnormalities of various safety indicators of Hebei Province.

"On the basis of the overall capacity of the Group’s Winter Olympic Safety Safety, we have established internal security service procedures to ensure that the support work is fine and do." Xie Shaofu said.

Artificial intelligence – depth learning – RNN’s defects and LSTM solution

1.1 RNN forward process

1.2 RNN reverse visual gradient process

BPTT (Back-Propagation Through Time) Algorithm is a commonly used method of training RNN, in fact, is also a BP algorithm, only RNN processing time series data, so based on time reverse propagation with time.

The central idea of ??BPTT is the same as the BP algorithm, and it is constantly looking for better points until the negative direction of the negative gradient direction that needs to be optimized. In summary, the BPTT algorithm is essentially the BP algorithm. The BP algorithm is essentially a gradient drop method, so the gradient of each parameter is the core of this algorithm.

It can be observed that at a moment of deflection on W or U, it is necessary to trace the information before this moment. This is just a minimum of a time, and it is also accumulated above, then the entire loss Functions will be very cumbersome of the biasing number of W and u. Although so but good, it is still tracking, we can write L of the W and u deflection numbers on the T time according to the above two oysters:

The overall bias formula is to add it again at all times.

As mentioned earlier, the activation function is nesting, if we put the activation function, take out the part of the middle:

Another expression method:

We will find that the multiplication will lead to the multiplication of the activation of the function derivative, which will cause the "gradient disappearance" and "gradient explosion".

1.3 gradient explosion (further a little more, N days, you will take off)

As long as the gradients in each stage are greater than 1, they will pass through N-wheels (such as n=365), and the last gradient will be very large.

The method of the gradient explosion should be simple:

As long as the gradient iteration is performed, the iterative gradient is checked. If it is greater than a value, the iteration is now in a certain value.

Note: The limitation here is: the value of the gradient during each iteration, not the value of the parameter W.

1.4 gradient disperse / disappearance (a little bit every day, N days, you will completely end)

As long as the gradients per level are less than 1, the N-wheels (such as n=365) are passed, and the last gradient will be 0.

1.5 RNN network gradient disappearance

This is a function diagram and derivative diagram of the Sigmoid function.

This is a function diagram and derivative diagram of the TANH function.

The two of them are similar, and the output is compressed within a range. Their derivative images are also very similar, we can observe that the derivative range of the Sigmoid function is (0,0.25], the derivative range of TACH functions is (0, 1], and their derivative is not more than 1.

This will result in a problem. In the process of multiplying the SIGMOID function as the activation function, then it is inevitable to make a multiplication, the less multiply. As time sequences continue to deepen, the decimal amount will lead to a gradient that is getting smaller and smaller until 0, which is the "gradient disappearance" phenomenon.

In fact, the RNN’s time series is very similar to the deep neural network. It uses the Sigmoid function in a relatively deep neural network to do activation functions, which will cause the gradient to disappear in reverse transmission. The gradient disappears means disappearing the parameters of that layer is no longer updated. Then the hidden layer turns a simple mapping layer, it is meaningless, so in the deep neural network, sometimes the number of neurons may be better than the increase in depth.

You may make an objection, RNN is different from deep neural networks, and RNN’s parameters are shared, and the gradient of a certain moment is accumulated at this time and before, even if it does not pass the deepest, the shallow layer is also gradient. This is of course pair, but if we update the more layers of shared parameters according to the gradient of the finite layer, there must be a problem, because the limited information is used as the optimization, and all information will not be found.

I have said that we have used Tanh functions as an activation function. The tanh function is also the biggest, and it is impossible to get all the values. It is equivalent to a pile of decimal in summary, or there will be "gradient disappearance. ", Why should I use it to do activation? The reason is that the Tanh function has a large gradient relative to the Sigmoid function, and the convergence speed is faster and the gradient disappears is slower.

One reason is that the Sigmoid function has a shortcoming, and the Sigmoid function output is not zero center symmetry. The output of Sigmoid is greater than 0, which makes the output are not 0 mean, referred to as offset, which will result in a signal that the neuron of the last layer will be output as an input of the non-0 mean signal. About the original symmetrical input and central symmetry output, the network will be better.

The characteristics of RNN were originally able to "chase trace" using historical data, and now tell me that the historical data available is limited, which is very uncomfortable, solving "gradient disappearance" is very necessary.

1.6 Solving the "gradient disappearance" method mainly:

(1) Select a better activation function

Generally, the RELU function is used as an activation function, and the image of the RELU function is:

The left orientation of the RELU function is 0, the right derivative is 1, which avoids the occurrence of "gradient disappearance".

However, the derivative of constant 1 is prone to "gradient explosion", but the appropriate threshold can solve this problem. Another point is that if the derivative of the left side of the left is likely to cause neurons to die, but the appropriate step size (learning brigade) can also effectively avoid the occurrence of this problem.

(2) Change the propagation structure

The LSTM network is the case of alleviating the gradient disappearance by changing the network structure.

1.7 RNN network function defect

(1) Unable to support long sequences

The RNN disappears due to the gradient, resulting in a time series of the RNN network, and therefore, the RNN network can only have short-term memories and cannot support the memory of the normal sequence.

This means that the RNN network may only process short sentences, unable to handle long text sequences composed of long text.

Although the RNN network structure seems to support long text.

This requires a new network structure to support long-term sequence input.

The so-called "long sequence" cannot be remembered, and only the context of "short", refers to the disappearance of the "long sequence" gradient.

"Braised ribs" appears in the beginning of the text, when entering the last string is, the RNN network, may have forgotten the most important word "red burn ribs" of this sequence, ….. The practice is similar to the hot chicken, finally It is possible to predict "hot chicken".

(2) The effect of the different words of the RNN network on the current output depends entirely on "time"

RNN network, at time is a series relationship, the more far from the remaining hidden layer, the smaller the output of the current hidden layer. RNN cannot affect the current output according to the importance of different words itself.

(3) RNN network is equal to all sequence inputs

RNN has the ability to extract all inputs with equivalent feature, and all words in a sentence will extract their features and as the input of the next feature. That is to say, the RRN network, remembers all information, is not distinguishing which is useful information, what is useless information, which is auxiliary information.

However, in the actual language, although different words are characterized, the role of the target is different, some words, in fact, there is no meaning, which is a modified word. The characteristics of some words, the role of decisive pairs, these words require memory within the scope of articles, that is, the importance of different words, not only depends on the physical distance between words, but also on The meaning of the word itself.

This requires a network that can discard and memorize the importance of the term.

Drop invalid information, remember that effective information is:

  • Enter the sequence can be growing, and the gradient disappearance is not easy.
  • Make effective information even if the current input distance is far from the current input distance, it can also produce a large influence.
  • Realize long-term memory

The network that meets the above functions is an upgrade version of the RNN network, the LSTM network.

2.1 LSTM Overview

LSTM (Long Short-Term Memory).

A long short-term memory network is a variant of RNN. The LSTM network combines short-term memory with long-term memory through exquisite door control, and solves the problem of gradient disappearance to some extent.

LSTM was raised by Hochreiter & SchmidHuber in 1997. Later, after being improved and promoted by Alex Graves, LSTM has achieved considerable success in many problems, so it has been wide in the scene entered by the sequence input. use.

LSTM achieves long-term memory functionality by deliberately designed basic components cell Cells.

2.2 RNN unit structure

In this configuration, there is only one memory state HT-1 that produces a new memory state HT together with the current input information.

The so-called memory, here is the value of HT-1, HT. The weight matrix of H correspondence is the extraction of memory.

The current input is XT. The weight matrix corresponding to the XT is instantaneous memory.

it’s here,

  • HT-1 is the memory of all historical inputs, that is, no discardment, no selective memory.
  • The farther history input is far from the current time, the smaller the current output HT, that is, long-term memory is forgotten.

HT=TANH (XT * WX + HT-1 * WH) is a history of history + instantaneous memory produces new memories.

2.3 Basic Composition Unit of LSTM Network: Cell

Remember that long-term information is the ability of the LSTM network natural, default, rather than the ability to pay large computing or memory costs. This is benefited from the LSTM unique Cell structure, that is, the basic composition unit of the LSTM network.

LSTM Cell is a memory unit with selective memory function.

(1) Trend of information flow:

  • This wheel timing information input: XT
  • Input merge: XT first merges with the current short memory HT-1, obtained (X, HT-1)
  • Forgotten door SIGMOD function switch: Through the WF matrix and (x, HT-1), determine if the output of this wheel needs to be used to remember, and how to use how to use how much long memory CT-1, the output of Sigmod is [0, 1]
  • Generate new information: The new information of this wheel is generated through the W matrix and (X, HT-1) and the TANH function.
  • Enter the door Sigmod switch: By Wi Matrix and (X, HT-1), it is indeed the new information generated ~ CT, superimposed to long memory CT-1, and how to use this new information, The output of Sigmod is between [0, 1].
  • Generate a new long memory: The long memory CT-1 at the last minute is superimposed with the new information ~ CT of this extracted to obtain CT=CT-1 + ~ CT.
  • Production of the output of this wheel: generates the output information of this wheel through the WO matrix and CT and Tanh functions, HT-1
  • Output door SIGMOD switch: By the WQ matrix and (x, HT-1), it is indeed necessary to output the output from the wheel to HT as a short time memory.
  • Timing information output of this wheel: HT and CT
  • (2) Two states of LSTM – external interface
  • RNN has only one hidden state (memory information) passes between adjacent units.
  • LSTM has two status information (memory information) to pass between two adjacent units.
  • A long time memory CT,
  • First, the short-term memory HT (instantaneous memory) at the last minute

(3) Special internal structure

There is only one additional operation and activation function inside the RNN.

The LSTM internal structure includes:

Sigmod activation function, play a role of analog switches, or called memory intensity, 0 means full forgotten, 1 means full memory, 0-1 represents partial memory.

  • Three multiplication: Multiplication is together with the Sigmod function, playing three "gate control". Phase 0 (close, forgotten), multiplied by 1 equal to input (open, memory).
  • A addition: Complete the current instantaneous input XT, short memory HT-1, long memory CT-1 output.
  • A Tanh activation function: The activation function required for the current period is output.

(4) Three doors of LSTM

  • Forget door, forgetting the door, the first Sigmod function: It decides that under the status of the current period and the current input, that is, this iterative output, you need to use the previous long-term memory.
  • Enter the door, which is the second Sigmod function: I decided how much the output of this iteration needs to be saved to the long-term memory of the small box, that is, what new content is needed. By entering the door, understand the input information of the current network.
  • Output door, but the third Sigmod function: Decided how to output information in the long-term memory small box.

2.4 Mathematical expressions of LSTM network composition unit

(1) The combination of XT and HT-1

[HT-1, XT]=Concatenate (HT-1, XT)

[HT-1, XT] is the input data generated by subsequent control doors and new information.

(2) Mathematical formula for forgetting the door

The Forcer Door is to indicate what kind of information we want to keep, what kind of information can pass, so the activation function here is Sigmoid.

(3) Enter the door

Establish an input door with Sigmoid to determine what value we will update, then build a candidate value vector with Tanh and add it to the state.

The above figure, in addition to the mathematical expression of the output door, the mathematical expression of this round of information is included in the new information ~ CT.

~ CT passes the gate, finally with long memory, CT-1 superposition (ADD), new long memory CT, 即 CT=CT-1 * forgotten gate control + ~ ct * input door control value, That is:

(4) Output door

The above figure, in addition to the mathematical expression of the output door, the final output HT of the wheel is further included.

2.5 Mathematical Expression

(1) Two inputs

[HT-1, XT]=Concatenate (HT-1, XT)

(2) Two outputs

(3) Intermediate unit

3.1 Series structure of RNN network

All RNNs have a chain form of repeating neural network modules. In standard RNN, this repeated module has only one very simple structure, such as a TANH layer.

3.2 Serial structure of LSTM network

LSTM is also such a structure, just repeated modules have a different structure.

Unlike a single neural network layer, the overall across H is flowing over time, and the cell state C is also flowing over time, and the cell state C represents long-term memories.

Copyright Notice: This article is the original article of CSDN blogger "Silicon Base Workshop", follows the CC 4.0 BY-SA copyright agreement, please attach the original link and this statement.

Do not scold the! Shenzhou solidarity association executives: We have to support more domestic brands like Lenovo

Wen | Northbridge proofreading |Northbridge

Without authorization, shall not be reproduced, copy or create images in any form!

Since the cult released a video about 7 Lenovo, Lenovo can be described as brought to the cusp of public opinion. After all, as an early associate of domestic technology companies, not only did the domestic chip, driven up operating system, we have become the mouth of the "assembly plant" Lenovo R & D investment accounted for less than 3% per year is a real ride "assembly plant "question. Lenovo executives Yang himself is released does not require self-developed chips with the operating system’s remarks, and users believe that such statements delayed the development of the domestic operating system.

It is also when Lenovo suffered users questioned, the founder Wu Haijun Shenzhou computer suddenly stand up and issued a solidarity association. Wu Haijun up to 1,500 words in solidarity among the nine major express opinions:

The first 1:00, founder Wu Haijun Hasee think Lenovo is the only one computer business will be the first global domestic companies, given an ulterior motive behind the upset, and the benefit will only be the second and third of Hewlett-Packard and Dell .

Point 2, Wu Haijun believe that the CAS year investment income of $ 200,000 would have 2.755 billion yuan, and is now owned by the Chinese Academy of Sciences also Lenovo shares worth 10 billion simply does not exist so-called loss of state assets.

Point 3, even if Lenovo executives accounted for half of the United States, but Lenovo has been done first in the world, indicating its restructuring is successful.

Point 4, Wu Haijun pointed out that a sales associate up to 4,000 billion yuan, although Lenovo executives got pay 10 billion, but sales only account for a fraction of point five of this compensation ratio is normal, even on the low side.

Point 5, the association has created numerous jobs, worthy of an enterprise play.

Point 6, although Lenovo liabilities 100 billion yuan, but Lenovo is just three months of sales, such debt is very healthy.

The first 7:00, although Lenovo sold abroad cheaper than domestic, but this brand is the need to support the home country of consumers. Wu Haijun and even Japanese brands, for example, many Japanese are Japanese brands only for those who support the Japanese brand to the world, and now Lenovo brand also needs the support of 1.4 billion Chinese people to be able to go further.

The first 8:00, I hope you see the situation, do not suppress domestic brands, the US brand advantage of.

Point 9, in which the existing computer market, the domestic PC market share of less than 50%, of which a large part is occupied by the association. If Lenovo fall, other domestic PC brands will be more difficult.

Although the Shenzhou Wu Haijun founder of Solidarity association issued thousands of words long article, but I still want to say that one of the loopholes or too much, it is difficult even netizens questioned.

For example, on the first 1:00, why at this time I felt Lenovo was the first company to achieve the world’s first business computer domestic brands, so that some people with ulterior motives questioned Lenovo is it? Why 5G standards of the time, the association did not vote for Huawei, but vote for the high-pass it? You know it is because of this critical vote, Huawei only missed the first. Well, this is it who have ulterior motives?

Furthermore Point 2, that someone pointed out that the year is not just Chinese Academy of Sciences has invested 200,000 yuan to Lenovo, Zhongguancun also offers a prime location, Wu Haijun is silent. There are 3 points now Lenovo executives have nearly half of Americans, so good on the Lenovo is a company to make people proud of the domestic brands do?

In point 4, the Wu Haijun think Lenovo executives to take pay 1 billion yuan is low, because Wu Haijun sales actually take the analogy, rather than net, you know Lenovo executives take pay ten accounted for net profit one points. Even if it is in accordance with the sales of Apple in 2020 to $ 600 billion in sales, Cook took $ 200 million in salary and stock dividends, this ratio is also lower than the remuneration of Association Executives.

As to point 7, is untenable ground, on what basis to support the needs of foreign markets by domestic consumers? Is domestic consumers is being taken advantage of it? Even so, the association also donate computers to the US military, which in turn is the motive of it? Look at the people Huawei, pricing is generally higher than the European market, Lenovo, Huawei Why would not think of such foreign selling it cheaper than domestic?

I want to say that today’s China cool with the Big Four already is not the original Chinese with a cool, once the association is not with today’s Huawei day difference, the more we need now is to focus on independent technology such as Huawei research and development of domestic brands, rather than eating procurement dividend company. Even support domestic brands, it should also be supported by more strength and potential of the company, this worthy of support domestic brands the consumer, rather than domestic brands to donate computers to the US military.

I ask you what is the matter think?

Users who know mobile phones, choose 256GB of memory when buying mobile phones, otherwise only yourself

Before the iPhone13 series released this year, there was a message that made Apple users were very unclear, that is, the iPhone13Pro series will cancel 256GB storage versions, because Apple’s previous mobile phone has this phenomenon, so many Apple users believe this information.

Everyone has such a big response because the 256GB version has become one of the standards of most mobile phone users when choosing mobile phone storage capacity. Therefore, more users will take a long time when buying a mobile phone, in fact, it is to store the size and high price.

I am buying iPhone13Pro Max, I started to choose 128GB version, because I have used 128GB version, from iPhone7plus, I have been 128GB version, but I feel that the storage space seems to be a bit not enough. Cancel the order to reselect 256GB version, the result is grabbed, and the version is not.

Then users are in the face of mobile phone storage 128GB, 256GB, 512GB, 1TB, so many capacity versions, what should I choose?

First, many users will have a misunderstanding, 128GB capacity is complete enough and cheap, 256GB is too expensive. Is it really?

Which volume version still has to look at the main purpose after the phone, if you start with a small size version of the price, it is likely to store space after a while.

The mobile phone is very weak, with the processor performance and camera technology, more users will use the mobile phone as a camera and entertainment tools.

I like to take photos, short video, 128GB is not enough, and the 265GB or even larger version is more secure.

In addition to storing photos video files, the App us usually uses a larger storage space. As we have more and more demand for mobile phones, the APP installed in the mobile phone is more and more, one mainstream mobile phone Just install hundreds of apps, and some of the large game apps occupy more than 10 gs.

In addition to the APP’s occupancy storage, the cache file that uses the app will also store space. For example, we use the most WeChat, my own WeChat cache has reached 27GB, accounting for more than 1GB of memory than 1GB. But you can’t clean up frequently, afraid that some data will be lost.

For users with low mobile phone-dependent, you can choose the stored version of your own, you will neither shoot a lot of photos and videos, and you will not play games, just use a mobile phone basic function, entertainment look at the news. , Video, these don’t have too much mobile phone storage, more wasteful.

Therefore, when the mobile phone users are replacing mobile phones, they are still based on their own scenarios and habits. If the annual memory occupancy is the same, they can keep the status, if their own mobile phone store has a growth trend, then choose a slight high A little storage version, after all, buy a mobile phone minimum to use 2-3 years, try to avoid the mobile phone because of insufficient storage.

[Little information] AI black technology piled up! This Baidu World Conference was stunned www.yidianzixun.com

In the video, Xiao Sang fell into Party A "true", and puts forward a lot of deceased demand for Baidu’s engineers. Not only wants this super robot to pick up their own travel, help yourself to run legs, and cultural literacy, will be poetry, Talk about life ideals, even when it is tired, it can turn into a rest space. This level of "super robot" sounds like a science fiction movie "delusion".

I didn’t expect this "secret weapon", just appeared at the Baidu World Conference yesterday. Baidu founders, chairman and CEO Li Yanhong really released "new species" – car robots.

There is no steering wheel and the accelerator pedal, which looks like a mobile space capsule, not only breaks through people’s imagination of the car, but also unveil the next chapter of the artificial intelligence spread.

Some people may have to ask, is it true that artificial intelligence really reaches this level?


Artificial intelligence is around

Artificial intelligence has long changed our daily lives.

It is no longer a computer and mobile phone. In the unknown, many people who are unpreparent in many people have been connected to the net.

Work on your body, glasses, screen, speaker, or even software in your mobile phone, TV on your TV, you can get out of a variety of things, find artificial intelligence shadow.

Smart life is no longer just a patented, artificial intelligence, making the "Science and Technology Gate Han" – the vulnerable group represented by the elderly, also across the "Digital Ditch", enjoy the welfare of scientific and technological progress.

Recently, "Beijing Seventy-year-old couple realizes self-service," is a typical example. Intelligent equipment is perfect in helping the elderly, solving the size of many people to solve the problem that many people need to help can solve.

Of course, this is not liberalization of children’s endowment obligations, and any equipment is more intelligent, and it is impossible to replace family companionship. The relationship between children and women’s busy and even wandering in the old and older groups, is also a social reality that cannot be ignored.

According to the seventh national census, the share of the 60-year-old and older population has reached 18.7%. According to the prediction of the Ministry of Civil Affairs, in the next five years, the elderly population is likely to exceed 300 million. my country is moving into the middle of the age.

Therefore, artificial intelligent equipment is of great significance in helping the elderly.

The importance of this meaning is reflected in artificial intelligence. Take the recent upgrade of small smart screen products as an example, Baidu enhances the "first aid" function and access free remote rescue services. When the user encounters a burst, for example, the old man falls in the house, just shouting "small degree, save me", Xiaomori smart assistant will not only call emergency contacts, but also directly involve professional rescue staff, Help and guide users to implement effective self-rescue and mutual assistance in the first time.

In addition to helping, in daily life, artificial intelligence has also played a huge role in improving the happiness of the elderly population. At the World General Assembly yesterday, Baidu announced a video, the popularity of CCTV, Wang Bingbing visited the four-story family of Beijing big fence, and felt the "black technology" of small house smart products. .

Using a small-scale smart product, users can interact with all house equipment by simple speech instructions. They automatically open the curtains after the guests will automatically open the fan and humidifier. Even the old home appliances can also achieve intelligence Transformation.

In addition to the functions present in the video, according to the recently released "Wufu Ai Hidden Program", Baidu’s wisdom, you can also help the elderly to enjoy health management, scientific exercise, body and body soothing, accompanying, community Waiting for the old-age service. The charm of "Black Technology" products is that technology can make older life richer, simpler, and more convenient.

Of course, the happiness increase in artificial intelligence spreads is also a popularity, and everyone can experience it.

Faced with various size challenges that have randomly appeared in life, if you can solve it quickly, the sense of accomplishment and satisfaction is often unpredictable.

At the World General Assembly yesterday, Baidu app spokesperson Gong Jun started from the camera knowledge map, voice search, booking booking, "Question" for help, and show us to quickly solve it in Baidu APP. What kind of experience is a kind of experience.

In order to solve the problem, we often need to search for filtering in the ocean in the vast information to find the answer. Now use the upgraded Baidu smart search, with the help of artificial intelligence, Baidu, the blade, perhaps gradually become our normal .

Today’s Baidu APP is no longer a simple search engine, but an integrated platform that uses artificial intelligence technology to make more people can more easily access knowledge and services in more occasions.


What is the black technology brought to society?

60 years ago, Dartmouth Meeting, marking artificial intelligence as an emerging discipline formally born.

At that time, the forerunters may have seen the economic potential of artificial intelligence applications, but more or for a kind of feelings and vision.

When this vision, crossing the long river of time, finalizing into a technological innovation, even if this innovation does not make money, it can also produce huge power.

For a few days, an innovative black technology called "Time Machine" on Baidu map has triggered a hot discussion on social platform.

Relying on the data accumulation of billionary positive maps, people can find pictures of different times at the Baidu map, you can feel the new day of China’s urban and rural construction, or you may find the previous memories in the time.

The shake is a 12 second video, 41 million points, saying is through Baidu map Time machine, let this blogger returns to the streets a few years ago, and see yourself through the time.

In order to achieve this function, it is necessary to spend huge cloud storage and cloud computing costs, while the image is acquired, including a variety of professional equipment, including high-precision map collection, also takes a lot of R & D costs.

However, such a project actually does not produce how much economic benefits, more is a public welfare nature.

But measure the value of a technology investment, in addition to economic benefits, it depends on the value and warmth of it to people and the society, and some technologies can achieve romance and warmth, and it can also become the original power of developers’ continued investment.

As a pioneer and explorer in the field of artificial intelligence, Baidu is transforming the technical achievements of these years, transforming into actions to fulfill social value.

In Quanzhou, Fujian, Baidu uses artificial intelligence to improve water resources management, and build a highly automated and intelligent "water brain" for Quanzhou.

The "Water Card" passes a few large screens, and it can perform real-time detection and regulation of the tap water pipe network throughout Quanzhou City. It not only reduces the risk of explosive tubes, but also has been greatly optimized, but only two or three can manage A tap water plant, but also guarantees the level of tap water to the pure water. Quanzhou citizens’ water safety and sustainable development of Quanzhou’s natural environment have been strong.

The application of "Quanzhou Water Card" helps Quanzhou Water Group’s overall personnel efficiency increased by more than 5%, and the energy consumption of water supply water supply unit decreased by 8%. The reliability of decentralized sewage treatment facilities increased by 5%, achieving good economic benefits, management Benefits and social benefits.

In Yunnan Ancient City Lijiang, Baidu is based on the infrastructure of artificial intelligence, and joints the Lijiang Municipal Government to create a set "Wisdom Lijiang City Brain", which is currently in Lijiang’s party building government affairs, cultural tourism, social governance, ecology Environmental, public service and other fields have played a huge utility.

In the party’s party government affairs, through large data application analysis model, Lijiang’s party and government integration platform has been stationed in 340 mainstream convenience services, and the time limit is compressed from 20 days to 3 days, and it has already completed more than 230,000 pieces. It is basically realized. "Villagers can’t do out the village."

In terms of social governance, through access to camera monitoring and artificial intelligence identification, Lijiang brain has automatically discovered more than ten cities in the row of garbage, motor vehicles, river invasion, violation operations, etc. The event is automatically identified, reported and closed, and the accumulated intelligent identification of more than 6,000 incidents have effectively improved the efficiency of urban governance and reduced management costs.


Technology has a fixed force, and the industry has grown

In this long change, you can’t leave the hardships of each pioneer, you can’t leave every entrepreneur’s initiative. Standing on the shoulders of countless ancestors, after the endless question and self-doubt, there were a few technologies such as Baidu’s full stack of AI capabilities, and took out the technical achievements sufficient to proud.

The report published by IDC shows that Baidu Smart Cloud has ranked first in the AI ??Cloud market for three consecutive times and ranked first in multiple sub-segment. As of last year, according to the cumulative pull request, Baidu’s depth learning framework "aircraft" is ranked second.

Behind the excellent rankings are the condensed strength of all people, knowing this Baidu, starting from a small grade, strengthening a road to open a win-win and giving back to society.

At the 2016 Baidu World Conference, Baidu first exhibited Baidu’s first artificial intelligence results – "Baidu Brain", and announced the core capabilities and underlying technology of the majority of developers, entrepreneurs and traditional enterprises to open Baidu brain. Since then, every Baidu World Conference is almost all surrounded by artificial intelligence. The more and more open pace, and the results of harvest are increasing. At present, Baidu Availability has learned an open source open platform and has helped more than 36 million developers.

After many years of deep cultivation, Baidu finally took a "big trick" at this year’s World Conference – "Car Robot". It is not only a collection of integration in Baidu’s huge investment in the past decade, but also the key to the 10-year gate of Baidu next AI.

In June last year, Baidu announced that in the next decade will continue to increase investment in new infrastructure in artificial intelligence, chip, cloud computing, data center. At the same time, Baidu also shows that the next five years is expected to continue to cultivate 5 million AI talents, providing AI talent guarantee for the development of intelligent economic society.

Over the past decade, Baidu’s total R & D expenditure accounted for more than 15%, such a continuous investment, not only much higher than the domestic Internet company, even if the foreign giants such as Microsoft, Amazon, is not inferior .

Long-term firm investment, bringing a fruitful technology accumulation. According to the "2020 Artificial Intelligent China Patent Technical Analysis Report" released by the Ministry of Industry and Information Technology, Baidu has been in the first three consecutive years of patent analysis reports in the artificial intelligence, and the number of patent applications and patent licensions last year reached 9364 and 2682 respectively. Part.

Behind these accumulation is a silent effort in Baidu for ten years, which is a hard exploration of thousands of multi-million developers day and night. It is also because of this persistence and strength, there is now the Baidu of "Second Entrepreneurship, Secondary Departure", and now we will continue to live in artificial intelligence.

What is the future of artificial intelligence? Perhaps the link between the AI ??chip and the brain, perhaps it is a human and robot, we still don’t know now.

But we know, Li Yanhong’s sentence is enough to reveal the upcoming "Tomorrow":

"In the future, AI will help our entire society, this is the power of the whole society from a larger perspective, and this is the power of Baidu to do AI – let people feel the power of technology.

It is recommended to praise it! It is more than 300 from the media to do more than 300, sharing 27 tools.

It is recommended to pay first to collect it! It is more than 300 from the media to do more than 300, share 27 tools, save it, don’t you find it for a while.

A good auxiliary tool can effectively improve our creative efficiency, or help us solve many problems we can’t solve.

Point to continue to see ~

First, material

1, sound tool:

(1) News is fast

(2) Wonderful W

(3) Baidu text

2, music material:

(1) Netease

(2) Cool dog

(3) Ear listen W

(4) Love to W

3, video material:

(1) Korean drama TV

(2) 80S

(3) Watermelon

Second, the tool class

1, editing tool:

(1) Trimming

(2) PR

(3) Fast clip

2, typesetting tool:

(1) 91 editing

(2) Xiimi

(3) 135 typography

3, font:

(1) Seeking font

(2) Webmaster material

(3) Chinese font

4, picture tool:

(1) Design

(2) Creatant paste

(3) Figure monster

Third, essential material

1, easy to use material W:

(1) Plum Wa W

(2) City of the sky

(3) Copywriting dog

2, picture material:

(1) element

(2) Thousand diagraph W

(3) Petal W

Many of us think that there is no problem, a software, tools, and websites can help us solve it. Is there a method skills and tools, and do you still do it in your media?

If there is anything about the media from the media, you can give a message to the big week.

Pay attention to the big week, you can learn a self-media dry goods that can make money every day.

What are the main applications of color stainless steel pipes?

Colored Stainless Steel Decorators are widely used in high-end villas, star hotels, advanced clubs, luxury cells, etc. At present, mature products have mainly screen, stair handrails, balcony guardrails, fences, gates, etc. It has the characteristics of color gorgeous, wear-resistant, and no fading. The surface treatment process has the following five types: wood grain transfer, vacuum electroplating, paint, spray, copper plated.

Stainless steel decorative tubes are mainly divided into 201, 304 tubes and 316 tubes. According to the type, it can be divided into ordinary tubes, tubes, and color tubes. For custom requirements, it is generally required for specific pipe diameters, elneins, flats, zoom, bending, etc., more for building decoration products, civil decorative art stainless steel products.

The color stainless steel tube is usually needed to do anti-fingerprint processing when it is used in public, or when it can be easily touched by hand; the anti-fingerprint treatment is to form a very thin and sturdy protective layer in stainless steel surface through nano-coating technology. Process, so that the stainless steel surface can not only achieve the effect of anti-fingerprint, but also improved corrosion resistance. Greatly improve the beauty and durability of colored stainless steel products.

"Network Security Challenge and Related Suggestions" in Automatic Driving

In February this year, the EU Network Security Bureau (ENISA) released the "Network Security Challenge" in automatic driving. The CCID Tank Network Security Institute compiled the report and expects to help our relevant departments.


The report combed the application status of artificial intelligence technology in automatic driving vehicles, and analyzed the network security threats and challenges generated in automatic driving through five attack scenarios, and proposed to strengthen the safety of artificial intelligence network safety, development End-to-end artificial intelligent security solutions, enhances the ability of artificial intelligence related events and vulnerability discovery capabilities, strengthening various countermeasures such as artificial intelligence safety training in the automotive industry. The report is designed to improve the awareness of potential risks of artificial intelligence technology and effectively resolve risks.


A new generation of cars are achieving semi-automatic and automatic driving capabilities with advances in artificial intelligence technology. According to the "Classification and Definition of Road Motor Vehicle Driving Automation System" (hereinafter referred to as SAE J3016) issued by the US Automotive Engineering Society (hereinafter referred to as SAE J3016), the road motor vehicle is divided into six automatic driving levels, from no automatic driving 0-level to complete automatic Driving and does not require the driver level 5, as shown in Figure 1. This report refers to the automatic driving vehicle, corresponding to 4 and 5-level vehicles in SAE J3016.

I. Application of artificial intelligence technology in automatic driving

(1) Artificial intelligence technology in automatic driving

In the past 10 years, the automatic driving is rapidly developed. The automatic driving system can make safety and smoothness of the destination by perceive and infer the surrounding environment, and take action to control the vehicle accordingly. The rapid development of artificial intelligence technology, especially machine learning is an important promotion of automatic driving. Figure 2 lists typical scenes facing the automatic driving system.

The automatic driving system enhancement function realized by artificial intelligence and machine learning technology mainly includes brake assist, smart parking, voice interaction with information entertainment system.

(2) Artificial intelligence software in the automatic driving system

Driving vehicles in a reality environment is not easy, requires complex social ethics and decision-making capacity to effectively respond to various accidents and hazards. Artificial intelligence software embedded in automatic driving vehicles can implement these functions, which are handled by various data collected by the processing sensor, thereby making decisions such as mobile, parking, and deceleration.

(3) Correspondence between automobile functions and artificial intelligence technology

Table 1 lists the correspondence between automatic driving vehicle function and artificial intelligence technology.

Second, artificial intelligence network security issues in automatic driving

The development of automatic driving and car networking has put forward higher requirements for the calculation function and interconnection of the vehicle, and also increases the possibility of a vehicle being attacked by a network. The network security risks of automatic driving vehicles will affect passengers, pedestrians, other vehicles and related infrastructure safety, and it is urgent to study the risk of safe vulnerability in application artificial intelligence. Artificial intelligent network threats can be divided into two categories: intentional threats and unintentional threats.

Common network security threats include four categories:

First yetSensor cents, blind, deception / saturation. The sensor will blind or disturbed. In this way, the attacker can manipulate the artificial intelligence model, provide error data to the artificial intelligence algorithm, or deliberately providing incomplete data, reducing the effectiveness of automatic decision making.

two isDOS / DDoS attack. Interrupt the communication channel available to automatically drive the vehicle so that it is basically lost in the outside world. This kind of attack can directly affect the vehicle operation and will prevent automatic driving. The purpose of the DDOS attack is to interrupt the communication channel.

ThreeManipulate the vehicle communication. The hijacking and manipulating communication channel will have a serious impact on the automatic driving operation, and the attacker can modify the transmission of the sensor reading or error interpretation of the information provided by the road infrastructure.

Four isInformation leakage. The vehicle in order to achieve automatic driving, which stores and uses a large number of personal sensitive information, including critical data saved on the artificial intelligence component, so potential attackers will intend to obtain such information, resulting in data leakage.

The following is interpreted by 5 hypothetical scenes.

Attack scenario 1:Confrared disturbances for image processing models used for street sign identification and lane detection.

Attack Scene 2:Intermediary attacks on the planning module.

Attack Scene 3:Data poisoning attacks identified by parking signs.

Attack Scene 4:Intrusion OEM background server, large-scale deployment of malware attacks.

Attack Scene 5:Sensor / Communication Interference and Global Navigation Satellite System (GNSS) spoofing attacks.

Third, artificial intelligent network security challenges and related recommendations in automatic driving

(1) Safety verification for systematic intelligence models and data

Data is critical to constructing and verifying artificial intelligence systems and is the core of the machine learning model learning process. Automatic driving vehicles typically configure multiple sensors, collect millions of environment description data per second, which provide basic support for complex, dynamic manual intelligence models.

(2) Providing a supply chain challenge related to artificial intelligence network security

Supply chain security is the weight of network security. In the supply chain of artificial intelligence components, if there is no appropriate security policy and enough strategy, it will lead to a lack of elasticity and there is potential security vulnerabilities. The security issues in the artificial intelligent life cycle may bring security risks to the automotive supply chain.

(3) End-to-end plan for artificial intelligent network security and traditional network security

In order to ensure the safe operation of automatic driving vehicles, artificial intelligent security solutions should be promoted in automatic driving vehicles and combined with traditional network security to improve the safety of artificial intelligence systems. Automatic driving deepening to artificial intelligence technology, not only provides power for attackers to implement cyber attacks on the target intelligence algorithm, but also the consequences of successful implementation of network attacks are getting more serious.

(4) Event disposal and vulnerability discovery capabilities related to artificial intelligence

Although many enterprise network security teams know that there is a network security hazard, there is a network security hazard, but only when security events or discovery vulnerabilities occurs, people really realize the importance of security. Although there is a lot of publicity of security vulnerabilities, people’s network security awareness is still weak, especially in the security vulnerabilities related to artificial intelligence systems.

(5) The lack of artificial intelligence network security and expertise in the automotive industry

Due to the lack of expertise in artificial intelligent network security due to development and system designers, there is no safety test and code analysis of artificial intelligence components during the development process, which has not set a network security policy in advance, which causes attackers to easily Automatic driving vehicle artificial intelligence components are locked as attack targets. Therefore, artificial intelligence security issues usually adopt the method of post-remedies, and the form of security inserts is mainly in the form of safety vulnerabilities.

Translated from:

Cybersecurity Challenge in the Uptake of Artificial Intelligence In Autonomous Driving, February 2021 by European Union Agency Fo Cybersecurity

Note: Reprint, add white lists and business cooperation, please leave a message in the public number, you must not reprint it, you must not modify and steal the original text.

Xiaomi 12 listing date file? Or bring a new small screen flagship model

Yesterday evening, the digital blogger @ è is very ballad to disclose a screenshot known as the "Millet House Sales Guide", where the road map points to: Double 12 will open on December 1st, new release will file 12 Month. At the same time, @ è is very ballad to ensure the authenticity of this document, it seems that there is no accident, the new product that is mentioned here should be Xiaomi 12.

Combined with the previous multi-explosive news, the new Xiaomi 12 series will continue high quality screen specifications, using micro-curved screen, equipped with Qualcomm Next Generation top flagship chip, Xia8 GEN1 chip, will adopt the design of the three plexes of CPU, of which the super large core frequency is 3.09 GHz, the large core frequency is 2.4GHz, the small core frequency is 1.8GHz, and the rabbit rabbit will break through millions of people. However, if the publisher is determined to be 28, Xiaomi should be a dead.

It is worth mentioning that the Xiaomi 12 series may also bring a new small mobile phone – Xiaomi 12X, the machine is compared to the flagship model, or casts a lot of peripheral parameters, such as the front screen resolution is equipped. Not 2k, but 1080p.

Although the screen size is not small, it is benefited from a narrower border, and the size of its fuselage should not be too large, and there is a small meter of Xiaomi 6. In the core configuration, it is not particularly powerful, or will be equipped with the Snapdragon 870 processor. This is a subkey processor, which is why Xiaomi 12X is one of the reasons for a second flagship. However, the Snapdragon 870 processor is not bad, no matter what is daily use or playing games.