New turn iMessage into a communication platform that can compete with the Messenger from Facebook, and allow businesses to interact with customers, writes TechCrunch.
Business Chat will appear on iOS 11, of which entrepreneurs will be able to delegate part of their functions applications into iMessage. So to book seats on the plane or to consult with a specialist, the user will not need to leave the chat.
Business Chat will be integrated with other products to serve customers such as LivePerson, Salesforce, Nuance and Genesys.
Companies will be able to integrate all business processes in the program, say the developers. A new feature will allow customers to schedule appointments and make purchases in a Business Chat.
See also: Apple invented his box of pizza
In may, a similar tool introduced the microblogging service Twitter. The company has allowed entrepreneurs to communicate directly with customers, conduct surveys among them and recommend the products.
From the coming iPhone 8 waiting for a really drastic change. Judging by the rumors and leaks, it has to be the first Apple smartphone with an OLED display, it has virtually no part, and the fingerprint scanner will be integrated directly with the screen. All the parameters should be the iPhone 8 one of the most advanced phones on the market, with the exception of data transfer rate. According to recent information, Apple will be forced to abandon the use of important technology.
It is no secret that Apple often resorted to the use of multiple manufacturers to produce various components in their devices, be it displays, camera modules, memory, etc. Often, this does not affect the operation of devices, but in the past year, some iPhone 7 and iPhone 7 Plus transmit data on LTE at noticeably lower speed than other smartphones. It was connected using LTE modems from Qualcomm and Intel. The decision from the latter is inferior in its capabilities that Qualcomm offers.
It is reported that this year Apple will again use the services of Qualcomm and Intel. At that time, as Qualcomm already has a modem with support for data transfer speeds of up to 1 Gbit/s, Intel will not have time to give Apple in the required number of similar decisions.
Many major mobile operators have already promised to launch before the end of the year Gigabit LTE network, but if the rumors are true, the new iPhone models this option will not be supported.
A person can improve their health by going to Church. Sure American scientists from Vanderbilt University.
To receive the results of its theory, the experts conducted an experiment, inviting you to take part in it more than five thousand people. Then they were divided into two groups, the first of which continued to live a normal life, and the second regularly went to Church.
The second category of people after telling the experts about improving your health and Wellness. They become easier to tolerate stress and negative life situations. The participants of the first group, no positive changes could not boast.
According to researchers, standing in the temple, the people are relaxed and quite different look and analyze what is happening around. After a while they are not so sensitive to every stimuli, thereby maintaining personal health.
Experts in network security found that hackers gathering almost half of the users of the Network. The results of the study were published in the academic journal ToDay News Ufa.
According to experts, almost half of user information available to hackers. This is due to the fact that every second modern man is the device through which it enters the Network. It is through hacking the system security of the gadget, the scammers get access to personal data. For example, in the media published an article, which reported that the company producing household appliances Vizio agreed with us regulators to pay a certain amount to be able to monitor the users.
According to scientists, from the manufacturer this behavior is unacceptable, it shows disregard for their customers. In addition represent a threat to children's toys with access to the Network. Hackers can access them and listen to the conversations that are happening in the family. Experts also said that in this way the scammers are able to obtain personal information and use it for personal gain.
Earlier, the website "Current news" reported that hackers were able to break the service OneLogin.
They paint pictures, write poetry, drive cars and beat human at Go. They implement in their services the world's leading companies. Google has dedicated a block of neural networks and AI for the I/O 2017, 2017 at WWDC Apple talked about the plans for their use, Qualcomm and Facebook announced a joint work on machine learning — the main tool of their development. And look, in a few years, some especially smart grid samoubijtsa to a full AI, and all will come complete Skynet. About whether this is what neural networks are and how they work in this article.
Say at once: neural networks — a concept very extensive related to math, physics and even chemistry that intersect with the area of artificial intelligence, machine learning, and of course programming. In the study and development of this sector involved a huge number of people, and the process began long ago. Therefore, in order not to turn informative text in the textbook of the history and theory of neural networks, will have to compromise: to omit much of the detail, names and dates for the sake of understanding the essence of the question.
A brief history of
To say "the history of neural networks" would not be entirely true, because the human brain is actually a highly complex neural network, and its history is one of the aspects of evolution. The first experiments in artificial neural networks (hereafter Ann) belong to the 1940 years, and they were set with the purpose of modeling and studies of the human brain.
1943. The neuroscientist Warren McCulloch and neurolinguist Walter Pitts created the first working artificial neural network. Despite the primitiveness of the first Ann are the neurons which could only operate on binary numbers, so their potential was considered huge due to the learning opportunity.
1960. More than fifteen years were spent on the creation of the first neurocomputer, the perceptron or "mark-1" which was developed by psychologist and neuroscientist Frank Rosenblatt. Device using solar cells to recognize printed on the cards letters.
The first neurocomputer "mark-1"
1969. Scientists Marvin Minsky and Seymour Papert revealed significant limitations of artificial neural networks. In addition to basic lack of resources to solve really complex problems, the ins was unable to implement some simple logic functions such as exclusive OR.
1975. Created multi-layer neural network that can change the strategy of solving a particular problem, depending on the source and the incoming data. The development of ins has moved forward.
1982. A full-fledged two-way communication between neighboring neurons, which further expanded the capabilities of the ins. In fact, the only limitation remained resources of computers that were still too weak for any serious tasks.
1980-ies. NETtalk developed the first neural network widespread. In the network tasks consisted of the pronunciation of English letters in the word depending on the context of the neighboring letters. Based on it also studied the mechanism of neural network learning, not only artificial. In the following decades, the neural network was complicated, there were new types to solve different problems. Develop Internet — that is, was the accumulation of structured data, necessary for operation of the ins. And most important — growing performance of computers.
Today, a simple neural network is able to function on not the most powerful servers or even smartphones, performing tasks at the end of last century was having difficulties supercomputers.
The principle of work — on your toes
Today it is believed that the human brain consists of about 86 billion neurons, between which there are synaptic (roughly speaking — electric) connection. Artificial neural networks, even the most powerful and ambitious, are much poorer and therefore much less productive — they still expect at best a model of the brain, but not a full artificial brain.
A schematic image of neural connections in the human brain
The classic and most simple neural network, or perceptron, is very simple: there is a layer of neurons receptors which receive information from the outside. Depending on settings, they either transmit the signal further into the network or not. The next layer of neurons receives signals from the receptors (usually several), processes them in accordance with a given algorithm and if the result reaches a certain — threshold — value, pass information on, the output layer of neurons, which give the result.
Diagram of perceptron
Given the structure of ins — many neurons, divided in layers — any such network produces parallel computing. A certain sequence gives the ply, but is very conditional, and in fact niveliruya bilateral data exchange between neurons on different layers.
It is also important to note that each neuron has a so-called weight factor — in simple terms, the significance coefficient for neurons with which it is associated. This is what determines the most important function of ins — ability to self-learn.
The essence of learning the ins
First of all, what is learning? Roughly speaking, it is the ability to understand that the end result invalid, and change the action for the faithful in this and similar situations. As does a neural network? Just as neurons in the human brain: depending on the information received, the intensity of the synaptic connections between them can change. In the Ann, each neuron has a certain weight factor, which varies depending on the correctness/incorrectness of the result.
The correct answer is determined by the person or the classic (that is, not neuro) program based on training samples with labels (example: "It is a machine. It is not a machine"). Having a certain number of correct answers, the ins can give the correct result outside the training sample. Such machine learning is called supervised learning.
An example of the formation of the simplest selections of correct answers
In a simple Ann with minimum number of binary neurons such training is very time consuming, but the result is too small. So over the years the ins has become increasingly difficult: there is an additional hidden layers (each functional unit of the human brain there are six of them), and achievement in determining the future of the ins, began the neurons of the two-way communication. Such a neural network has been called recurrent, in which neurons "throws" information to each other several times, changing their weights, as long as the last layer does not give the correct answer. The correct answer is automatically adjusted based on the analysis of data training samples without any labels. It is unsupervised learning. When the neural network produces a result and then receives information about its correctness/incorrectness is reinforcement learning.
A simplified diagram of the multi-layer neural network with feedback neural communication
The choice of type of training of a particular Ann is determined by the task for which you created the network. Learning with a teacher is to define the objects in the pictures; without a teacher for structuring, organizing large amounts of data; reinforcement — to predict when the input data are constantly changing.
In addition to the classification method of the learning network are divided into different types according to the structure, nature of relations, type of input data and other characteristics.
For example, for image recognition using so-called convolutional network. The principle of their work is drawn from the principles of the visual cortex of the brain. The particular features of image the neural network moves to the more abstract parts, and further to more abstract details until the release of the high-level concepts. As an example consider a Ann that analyze data from a traffic surveillance camera. Her first task is to identify the car in the frame. Then the following tasks: the speed of the car, wearing a seat belt if the seat belt is not listed if the car was stolen (you need to count the number and compare with database), and so on. Optional: to define the color and make of the vehicle (if stolen), to find information about the driver, the owner. All this can make one well-trained convolutional neural network with sufficient number of resources.
A logical question arises: what is the General theoretically known to limit the development of neural networks, how smart can they be? We decided to ask expert on the ins.
"There are two serious limitations — the number of different training data and computing power on which to learn. Neural networks can solve problems with clear conditions and clear criteria of success. For example, in voice recognition there is a clear criterion of quality: the number of correctly recognized words. And in the task of writing a literary work, it is unclear how to evaluate progress, so this task for neural network learning is seen unsolvable", — the head of service vision of "Yandex" Alexander Krainov.
If you keep all areas in which today apply to the ins, get a decent encyclopedia. For example, in electronic viewfinder, face detection, smile detection, gesture, movement — all this is the work of the neural network. Based on the acclaimed app Prisma is the ins. Translation from a foreign language, definition of text on photo, voice assistants, drawing seals on the basis of the sketches and all... well, you understand. Ins are widely used in search engines, and unmanned vehicles.