My mother is a nurse and is currently retired. She is a very intelligent person and very responsible for her work. A few days ago, I told her that I was working on Imagination's latest neural network accelerator. She surprisedly said, "What do you mean?" Of course, only when she is undergoing surgical training in nursing school or caring for dementia patients Really think about what a neural network means. I explained to her that I was talking about artificial intelligence (AI), not the human brain. I decided to write an introduction to neural networks so that my mother (and other people without technical background) could have some knowledge of neural networks. Understand and use other reading materials to truly enter the world of artificial intelligence.
What is a neural network?
If you use Google Chrome to search and view "What is a neural network" (of course, this is a question frequently asked by everyone), you will find that there are many academic articles, each of which has a picture similar to the picture below And explain the function of neural network in detail, involving many professional terms such as "perceptron" and "back propagation", etc. If you are not good at mathematics or engineering, then you will feel bored reading a few paragraphs.
Although most of us are not professional engineers and may not be able to really understand how these algorithms work (I'm not even sure if those technical people really understand the mysteries-such as the black box problem), neural networks still have a lot of usefulness The information needs us to understand.
Because neural networks are constantly changing our lives! I'm not talking about "Skynet (Terminator Movie)" or VIKI (Supercomputer in the movie) etc. (but some scenes do happen in real life), I am talking about what we experience every day, some new technologies We are changing our way of life such as smart speakers / sound assist, shopping, photo editing, fraud detection, etc. There are also some technologies that are upending some industries, such as self-driving cars. There are also technologies that may dramatically change the global society (such as genomic medicine), and there are many changes that we have not yet conceived.
Ophir Tanz (@OphirTanz) and Cambron Carter published an article on the TechCrunch website, decomposing the question "how does a neural network work?", But in a broad sense, what is really missing is what we can understand Neural network terminology.
Break down the problem
What is the precise definition of artificial intelligence (AI) ? I simply checked the Oxford dictionary: AI is the result of computer system theory and development. It can perform tasks similar to human intelligence, such as visual perception, speech recognition, decision making, and language translation. AI is also divided into multiple subsets, including robotics, machine learning, natural language processing, etc. In 1956, a research on artificial intelligence (AI) was first proposed at a seminar at Dartmouth College, although I will not introduce it to you here. A piece of history, but encourage everyone to understand it by themselves.
What we know now is " Narrow AI ", which is divided into many types of networks, each of which faces a specific function (such as picture classification, etc.), and the continuous development of this science will eventually develop into " general artificial Intelligent (General AI) , when the computer will act and think like humans, although we have not reached this level, but after that era, the computer can complete every intelligent work of humans, including reprogramming itself ( (Similar to the "Skynet" and "VIKI supercomputer" in the movie, it may cause panic at that time). But many people are working hard to ensure that some scenes only exist in science fiction. You may wish to read @ cademetz's article: Teach the AI ​​system to regulate its own behavior.
Machine learning is currently the most advanced field of AI science. It describes how machines collect data, parse and make some predictions based on these data. The machine system will perform self-learning without manual programming. Only recently have we applied its powerful processing power to reality. Many large companies such as Google and Amazon are using machine learning technology.
Deep learning is a type of machine learning. It needs to design many layers in the network for data processing, which may reach tens of thousands of layers. For example, convolutional neural networks (CNNs) are very suitable for image detection and classification. It has designed dozens or hundreds of network processing layers, each of which is responsible for detecting some complex attributes, from learning how to detect edges to finally learning to detect images. As for the specific shape, the performance of deep learning networks will continue to improve with the continuous increase of training data.
Artificial neural networks (ANNs or NNs for short) are similar to CNNs networks and are also intelligent networks based on computer processors. They can work like the human brain (or close to the human brain). NNs networks use different mathematical algorithms to process the network layer, which can continuously improve the The sensitivity of the received information, from pictures to voice, text, etc. In addition, there are many types of neural networks. In addition to CNNs, there are BRNNs, DNNs, FFNNs, LSTMs, PNNs, RNNs, TDNNs, etc., which are not introduced in detail here. Each type of network faces different specific functions. If you want to learn more, a detailed introduction is made on the website of the Asimov research institute.
For neural networks (NNs) to be used in specific applications, a large amount of data needs to be input for training. For example, convolutional neural networks (CNNs) are used for image classification. The app for "identifying hot dogs" implemented by the CNN network can sort out pictures of hot dogs, sausages, etc. from 150,000 photos (really achieved!). @timanglade wrote a process for training this network: "How to use TensorFlow, Keras & React Native to identify hot dog App applications in the Silicon Valley TV series launched by HBO." To train this network to achieve high precision requires a lot of computing power and often requires the help of a large number of GPU arrays in the data center.
"Identify Hot Dog" App Application Prototype After training, the neural network can parse new input data and make certain predictions through "inference" operations. For example, in this "Hot Dog Identification" application, you can point the phone camera at an object and it will help you determine whether the object is a hot dog. In order to achieve this function, it is necessary to quickly make inference sorting based on the understanding function obtained by a large amount of training before, and finally provide the probability of whether an object is a hot dog.
Inference operations (not only for detecting hot dogs!) Are exactly what the PowerVR Series2NX neural network accelerator (NNA) is good at. Its performance is far more than that of mobile phones. We will witness its performance in drones, surveillance cameras, automobiles and other guides. According to the application environment in which it is located, we will provide new services to us based on certain inference operations. This 2NX network recognizes videos of different objects.
Neural network in application
As I mentioned before, different networks are good at doing different things. I do n’t agree with exaggerating a certain type of network (this is something that other websites often do). I think what is more important is what this network can achieve through training Features. The functions of the NNs network currently used include recognition, classification, image quality improvement, face and various object detection and recognition, etc. Neural networks (NNs) are increasingly used for language translation, music production, and text / speech recognition. "Alexa, are you using a neural network?" (Yes, she does).
Amazon, Google, Baidu, and Facebook all use neural network technology (NNs), and we also use these technologies on different platforms every day, from search engines to social media.
In practical terms, the algorithm used by Amazon makes our shopping easier, but there are also competing algorithms that often recommend me some unrelated books, so I may also miss a lot of books that I should read ( Advice from Goodreads website), I have to suspect that this is related to paid advertising, I hope they can take certain measures to make these malicious competition phenomena less obvious.
A book catalog recommended to authors on Amazon based on a neural network (NN) engine As a fair-skinned redhead with a family history of melanoma, I am also very happy to work hard to build a neural network system that can monitor skin cancer with high precision through photos.
If you feel boring and boring, you can try to use some App with neural network technology, use the mobile phone camera to scan your surroundings, can provide you with some unexpected information according to the picture.
Point your mobile phone camera at a person with red hair features, open the App application, there will be voice report some related information, you can even change the real-time photo screen, become younger or more happy Photographs of people's faces, but fake news cannot be created in this way!).
A blog written by Yaron Hadad provides a detailed overview of some amazing things that deep learning achieves, and there are some very cool videos.
Of course, for any technology, someone will abuse it, and the neural network will continue to process the feedback data. Whether it's password cracking, stealing money or identity, or hijacking infrastructure systems, we know that there are still many people who are "crazy" to study these.
Looking to the future
Finally, I hope my mother can have a basic understanding of neural networks (NNs), and like me, I will continue to be inspired by the research results of researchers and use these technologies. As we come into contact with more data, better and more economical processing technology, such as PowerVR 2NX neural network accelerator, deep learning technology, etc. will be integrated into every field of our society. They will be used in self-driving cars, managing smart homes and offices, significantly increasing productivity, keeping everyone away from hospitals, and providing endless fun for everyone.
With 15+ years manufacturing experience for Portable Power.
Supply various portable charger for iPhone, Airpods, laptop, radio-controlled aircraft ,laptop, car and medical device mobile device, ect.
From the original ordinary portable power source to Wireless Power Bank, Green Energy Solar Power Bank, Magnetic Mobile Power, Portable Power Stations and other products continue to innovate.
Avoiding your devices run out of charge, Portable Chargers to keep your mobile device going.
We help 200+ customers create a custom mobile power bank design for various industries.
Wireless Earphones,Wireless Headphones,Best Wireless Headphones,Best Bluetooth Earphones
TOPNOTCH INTERNATIONAL GROUP LIMITED , https://www.micbluetooth.com