Facebook official blog update, FAIR executive, deep learning representative Yann LeCun and colleagues write an article to explain what artificial intelligence, artificial intelligence affects our lives, and how we will learn, work and live in the future full of artificial intelligence. Facebook also launched a series of instructional videos to help you better understand artificial intelligence.
Tuesday morning at 8:00 am. You are already awake, glanced at the title on the phone, replied to an online post, ordered a holiday sweater for your mother, locked the house and drove to work, and listened to some nice songs on the road.
In the process, you have used artificial intelligence (AI) a dozen times - wake up by an alarm, get local weather reports, buy gifts, lock your house, learn to remind you of upcoming traffic jams, and even identify one The first unfamiliar song.
AI has spread all over our world, and it has undergone tremendous changes in everyday life. But this is not the AI ​​you saw in science fiction movies, nor the nervous scientists slamming the keyboard, trying to stop the machine from destroying the world.
Your smartphone, house, bank and car are already using AI every day. Sometimes it's obvious, just like when you let Siri direct you to the nearest gas station, or Facebook suggests that you remind a friend that you posted an image online. Sometimes it's barely visible, just like when you use your Amazon Echo to buy something you don't usually buy on your credit card (such as a fancy holiday sweater) and don't get a fraudulent SMS alert from the bank.
AI will bring about a major social transformation by promoting the development of autonomous vehicles, improving medical image analysis, promoting better medical diagnosis and personalized medicine. AI will also be the underlying framework that will support many of the most innovative applications and services of the future. But for many people, AI is still mysterious.
To help you unlock these puzzles, Facebook is creating a series of educational online videos outlining how AI works. We hope that these brief introductions will help you understand how the complex computer science field works.
Not magic, just code
First of all, there are some important things to know: AI is a rigorous science that focuses on the design of intelligent systems and intelligent machines. The algorithmic techniques used in this way draw on our understanding of the brain to some extent. Many modern AI systems use artificial neural networks and computer code to simulate a very simple network of interconnected cells, a bit like neurons in the brain. These networks can learn from the experience by modifying the connections between the units, a bit like the human and animal brains learning by modifying the connections between neurons. Modern neural networks can learn to recognize patterns, translate languages, learn simple logical reasoning, and even create images and form new ideas. Among them, pattern recognition is a particularly important function - AI is very good at identifying patterns in large amounts of data, which is not so easy for humans.
All of this happens at an alarming rate through a set of coding programs, and the neural network running these programs has millions of units and billions of connections. Intelligence comes from the interaction between these many simple elements.
Artificial intelligence is not magic, but we have seen how it can dramatically advance scientific research like magic, and play an important role in identifying objects, recognizing speech, driving cars, or everyday miracles that translate online articles into dozens of languages. .
At the Facebook Artificial Intelligence Research (FAIR) lab, we are working hard to make learning machines work better. A large part of it is called deep learning. Using deep learning, we can help AI learn the abstract representation of the world. Deep learning can help improve speech and object recognition and help advance research in physics, engineering, biology, and medicine.
A particularly useful architecture in deep learning systems is called convolutional neural networks or ConvNet. ConvNet is a specific way of connecting cells in a neural network, inspired by other animal and human visual cortex architectures. Modern ConNet can utilize units from 7 to 100 layers. In the park, we humans see the Collie and Chihuahua. Although they are different in body size and weight, we know that they are all dogs. For computers, images are just a bunch of arrays. Within this array, local patterns, such as the edges of objects, can be easily detected in the first layer. The next layer of the neural network will detect the simple shape of the combination of these simple patterns, such as the wheel of a car or the eye of a human face. The next layer will detect certain parts of the object formed by these combination of shapes, such as the face, legs or the wing of an airplane. The last layer of the neural network will detect the combination of those parts: a car, an airplane, a person, a dog, and so on. The depth of the neural network—how many layers—allows the network to recognize complex patterns in this hierarchical manner.
Once trained in a large sample database, ConvNet is especially useful for identifying natural signals such as images, video, speech, music, and even text. In order to train the network well, we need to provide a large amount of image data that is marked by these networks. ConvNet will learn to associate each image with its corresponding label. Interestingly, ConvNet can also pair images and their corresponding tags that have never been seen before. So we got a system that can sort through a variety of images and identify the elements in the photo. These networks are also very useful in speech recognition and text recognition, and are also a key component in autonomous vehicles and the latest generation of medical image analysis systems.
What can be learned
AI also solves one of the core problems that we humans face: What is intelligence? Philosophers and scientists have been working hard to solve this problem, and the answer has been elusive and erratic, even if this center is the fundamental attribute we can call people.
At the same time, AI also raised a lot of philosophical and theoretical questions: What can be learned? The mathematical theorem tells us that a single machine that can learn cannot effectively learn all the possible tasks, and we also know what is impossible to learn, no matter how much resources you invest.
In this way, the AI ​​machine is like our human being. In many ways, we are no better than machines that we will learn. The human brain is highly specialized, albeit with obvious adaptability. The current AI system is still far from having the seemingly general intelligence that humans have.
In AI, we usually consider three types of learning:
Reinforcement Learning is about how agents should act to maximize rewards. It is inspired by the theory of behavioral psychology. In certain situations, the machine picks an action or series of actions and gets a reward. Reinforcement learning is often used to teach machines to play games and win games, such as chess, backgammon, go or simple video games. The problem with intensive learning is that simply strengthening learning requires a lot of trial and error to learn simple tasks.
Supervised learning Basically, supervised learning is the correct answer we tell the machine to input: this is an image of a car, the correct answer is "car." It is called supervised learning because the algorithm learns from tagged data in a process similar to showing a picture book to a young child. Adults know the correct answer and the child makes predictions based on the previous examples. This is also the most commonly used technique for training neural networks and other machine learning architectures. For example: give a description of the large number of houses in your city and their prices, try to predict the price of your own house.
Unsupervised learning Humans and most other animals learn in the first few hours, days, months and years before their lives, in a way that no one supervises: we understand by observing and knowing the results of our actions How the world works. No one tells us the name and function of every object we see. We learn very basic concepts. For example, the world is three-dimensional, objects do not disappear by themselves, and unsupported objects will fall. At the moment we don't know how to achieve this on machines, at least not at the level of humans and other animals. The lack of AI technology for unsupervised or predictive learning is one of the reasons for limiting the current development of AI.
This is a method that AI is often used, but for any computing device, there are a lot of problems that are fundamentally unsolvable. That's why even though we built machines that transcend human intelligence, these machines still have limited capabilities. These machines may beat us when we play chess, but we don't know how to hide in the house when it rains.
future career
As AI, machine learning and intelligent robots become more common, these robots will take on new positions in manufacturing, training, sales, repair and fleet management. Artificial intelligence and robots will be able to implement new services that are unimaginable today. But it is clear that health care and transportation will be the first subversive industry of AI.
Young people can enjoy the many opportunities offered by AI as long as they adjust their career goals. So how do we prepare for a job that doesn't exist yet?
If you are a student:
Mathematics and physics are places to learn artificial intelligence, machine learning, data science, and many of the basic methods of future work. Take all the math courses you can take, including Calc I, Calc II, Calc III, Linear Algebra, Probability and Statistics. Computer science is also essential, you need to learn how to program. Engineering, economics, and neuroscience can also help. You can also consider some areas related to philosophy, such as epistemology - this study studies what is knowledge, what is scientific theory, and what is learning.
The goal of taking these courses is not a simple memory. As a student, you must learn how to turn data into knowledge. This includes basic statistics, including how to collect and analyze the data, paying attention to possible deviations, and being careful about the errors that occur when processing the data.
Ask your school's professor that he or she can help you and make your ideas more specific. If their time is limited, you can also ask a senior doctoral or postdoctoral.
Read a doctor. Don't worry about the “ranking†of the school, find a reputable professor in a research that interests you, or choose someone who writes a paper you like or admire. Apply for some doctoral programs at the schools where these professors are located, and mention in the application letter that you are willing to work with these professors, but are also willing to work with others.
Participate in researching AI-related questions that you are interested in. Start reading the literature on this issue and try to solve it differently than before. Before you graduate, try to write a research paper or post an open source code.
Apply for internship opportunities that focus on industry and gain experience in AI in practice.
If you are already employed, but want to switch to work related to AI:
There is a wealth of information on in-depth learning online, including lectures, online materials, tutorials and machine learning related courses. You can sign up for the Udacity or Coursera course, read Yoshua Bengio, Geoff Hinton and my co-authored Nature thesis, and the recently published book Deep Learning, by Goodfellow, Bengio and Courville, and my recent French Academy in Paris. Lecture (in English).
Of course, you can also consider returning to study, then refer to what I said above.
Looking to the future
More and more human intelligence activities will be carried out with intelligent machines. Our wisdom is the foundation of our becoming a human being, and AI is an extension of this property.
On the road to creating truly intelligent machines, we are discovering new theories, new principles, new methods, and new algorithms that will produce applications that will improve our daily lives today, tomorrow, and beyond. Many of these technologies are quickly being used for Facebook products and services, such as image recognition, natural language understanding, and more.
When it comes to Facebook AI, we have a long-term goal: to understand intelligence and build smart machines. This is not just a technical challenge, it is a scientific issue. What is intelligence, how do we reproduce it in the machine? In the end, this will be a problem for all mankind. The answers to these questions will help us not only build intelligent machines, but also gain a deeper understanding of mysterious human thoughts and how the brain works. If possible, these answers will also help us better understand why human beings are human.
Smd Buzzer,Smd Piezo Buzzer,Smd Micro Buzzer,Smd Magnetic Buzzer
NINGBO SANCO ELECTRONICS CO., LTD. , https://www.sancobuzzer.com