R&D personnel have a dream to let artificial intelligence machines create other artificial intelligence machines. However, this is a nightmare for senior programmers.
Recently, in the speeches of Silicon Valley and China, Jeff Dean, one of Google's chief engineers, mentioned AutoML. ML, the acronym for machine learning, is a computer algorithm that can perform specific tasks by analyzing data. AutoML is also a machine learning algorithm, but it learns how to create other machine learning algorithms.
With AutoML, soon, Google may find a technology to create AI that partially replaces humans. This is the future of the technology industry, many people think.
From smart phone applications that can recognize faces to auto-driving cars, the technology industry has bright prospects. However, it is estimated that there are only about 10,000 people worldwide who have the educational background, experience, and ability to create complex (even somewhat mysterious) algorithms.
The world’s largest high-tech companies, including Google, Facebook and Microsoft, invest millions of dollars each year in AI talent, making this scarce talent market even more difficult. For hundreds of years, talent shortages are difficult to solve in a short period of time, and it takes several years to master these skills.
However, industrial development will not wait. At present, companies are developing various tools to help simplify the process of creating AI software, such as image and speech recognition services, online chat robots, and so on.
Recently, Microsoft announced a tool that can help programmers create deep neural networks. "Every new technology emerges, and computer science will follow suit in this way," said Joseph Sirosh, Microsoft's vice president. "We are saving many laborious tasks."
This is not altruism.
Researchers like Jeff Dean believe that more people and companies studying artificial intelligence will promote their own research. The giants also saw a bright "money" path in this trend. They are selling cloud computing services to help companies and developers create AI.
"There is a need for this, and the current tools cannot fully meet these needs," said Matt Scott, co-founder and chief technology officer of CodeLong Technologies. This startup company is also committed to similar services.
This is exactly what Google wants AutoML to do. Google CEOOSundarPichai also showed off AutoML when it released a new Android smartphone last month.
This project will eventually help other companies create systems with artificial intelligence, even if these companies do not have a wealth of expertise. Jeff Dean said that only a few thousand companies now have the ability to create AI, more companies have the necessary data.
"I hope we can solve the machine learning problems of thousands of organizations first, and then help millions of company organizations." He said.
Google is investing huge amounts of money in its cloud computing business and expects it to become the company’s main economic engine in the coming years. After putting most of the world's top AI R&D personnel in the bag, Google also has the weight to start the engine.
Neural networks are rapidly accelerating the development of artificial intelligence. Today, engineers can create autonomous learning algorithms faster. In the past, programmers manually created an image recognition service or a text translation app that could only complete one line at a time.
For example, machine learning algorithms can learn to recognize spoken words by analyzing a large number of traditional techniques to support the voice in a call.
However, building a neural network is not as simple as creating a website or an ordinary smartphone application. Mathematics, extreme trial and error, quite a lot of intuition, one can not be less. Jean-François Gagné, CEO of ElementAI, an independent machine learning lab, defines this process as "a new type of computer programming."
In the process of creating a neural network, R&D personnel often use a large number of machine networks to run dozens or hundreds of experiments to test how much the algorithm can learn to recognize images or translate a language. They will constantly adjust specific parts of the algorithm until they find a part that works. This technique is called "dark art" because R&D personnel find it difficult to explain why they made such adjustments.
Google is trying to automate this process using AutoML. By creating algorithms that can analyze the development process of other algorithms, you can figure out which methods work and which do not. Ultimately, the system will learn to create more efficient machine learning.
Now, Google believes that AutoML can create algorithms that recognize picture objects more precisely than algorithms built entirely by human experts. Barret Zoph, a member of the project and Google researcher, believes that this approach will eventually apply to other tasks such as speech recognition and machine translation.
This is one of the important trends in AI research. Experts call it "learning to learn" or "meta learning."
Many people believe that such an approach will greatly promote the development of artificial intelligence, both in the online world and in the real world. Berkeley's R&D personnel are building technologies that allow robots to solve new problems based on past experience.
"Actually, computers will create such algorithms for us," said Berkeley's professor Pieterabbeel. "Computer-generated algorithms can solve many problems quickly. At least we have such hope." Recently, he left OpenAI and set up a company. Robot startup company. (The other general will leave the open AI portal in Musk, saying that the industrial robot market has promising prospects)
This is also how more people and businesses can build AI. These methods will not completely replace human researchers. Experts still need to do a lot of important design work. However, they hope that the results of the work of a few experts can help more people create their own software.
It is still not realistic. It will be realized in the coming years. It is only a matter of time. RenatoNegrinho said. He is a researcher at Carnegie Mellon University and is also developing technologies like AutoML.
PCB Screw Terminal Block section
PCB Screw Terminal Block
Screw type PCB wiring terminal, the structural design takes into account the characteristics of convenient wiring and solid screw connection. It can connect all types of conductors with a cross section of 0.2mm to 35mm, with a spacing of 2.5mm-19.0mm.
The screw type terminal block has the advantages of convenient connection and firm screw connection. It can connect all types of conductors with cross-section of 0.08mm2 to 25mm2, and the spacing is 2.54mm to 15.0mm. The product design conforms to iec60998, ul1059 CSA c22.2 No.158 and other international standards. ]
PCB Screw Terminal Block
Screw type PCB wiring terminal, the structural design takes into account the characteristics of convenient wiring and solid screw connection. It can connect all types of conductors with a cross section of 0.2mm to 35mm, with a spacing of 2.5mm-19.0mm.
The screw type terminal block has the advantages of convenient connection and firm screw connection. It can connect all types of conductors with cross-section of 0.08mm2 to 25mm2, and the spacing is 2.54mm to 15.0mm. The product design conforms to iec60998, ul1059 CSA c22.2 No.158 and other international standards. ]
PCB Screw Terminal Block,PCB terminal blocks,terminal blocks,industrial terminal bllocks,green terminal blocks
ShenZhen Antenk Electronics Co,Ltd , https://www.antenkconn.com