Shulou(Shulou.com)06/02 Report--

Produced by big data Digest

Compilation: Wang Yiding, Zhang Qiuyue, Luo ran, Qian Tianpei

In the winter of 1975, a bit of "strange" news appeared on the billboard on the San Francisco Peninsula.

"are you trying to do your own computer development? if so, come to our party!"

This announcement comes from the Homebrew computer Club of that year. Homebrew is an amateur community, and club members are interested in the potential of a new component called microprocessors at the time.

Homebrew's first meeting was held in a garage in Menlo Park, California, and was attended by 32 people, including a young engineer named Steve Wozniak. He later introduced his friend Steve Jobs to the club.

Len Shustek, a retired entrepreneur who attended the conference, said: "this club proves that not all technological advances have to happen in big companies and universities."

For artificial intelligence, the same thing is happening.

Thanks to artificial neural networks and related technologies, computers have become more and more capable of understanding voice and images since 2012.

To really master this kind of artificial intelligence technology requires a deep understanding of computers, many years of research experience and solid mathematical skills. If you have all these conditions, congratulations: giants like Amazon, Facebook, Google or a handful of other giants that actively embrace artificial intelligence strategies are already scrambling to grab you.

However, in this "AI first" competition, there is no shortage of tools and spare parts that anyone can learn and use. To attract top scientists and application developers, tech giants have released some internal AI build kits and some of their research for free.

In the process of realizing the Silicon Valley AI dream, professional and non-professional enthusiasts are using almost the same technology.

"now, high school students can do what the best artificial intelligence practitioners couldn't do a few years ago," said Wu Enda, who is in charge of projects related to Google and Baidu.

Represented by Wu Enda, many people think that a breakthrough in artificial intelligence in the field of non-traditional computers is very promising: they hope that it is not limited to Silicon Valley, but that it should widely spread the potential of technology in physics and culture. see what happens when technical laymen "train" neural networks according to their own preferences and observation methods.

Wu envisioned a scenario in which a person in India might one day make local drinking water safer by learning from online videos of artificial intelligence.

Of course, there are some DIY neural networks that are "not suitable for children". At the end of last year, a Reddit account posted what appeared to be a pornographic video of Wonder woman actor Gal Gadot, which was circulated on Reddit and later on an adult video site. But attentive viewers will notice that Gadot's face in the video occasionally appears like a loose mask.

The publisher explained that the clip was generated by training the neural network: a Gadot image that matches the facial expressions of the original character in the video. Then they posted the code and methods online so that anyone could make deepfake clips similar to their own.

All in all, what the DIY era of artificial intelligence brings to us may not be all positive, and certainly not all negative. In most cases, it will perform very well in specific areas.

Let's take a look at the pioneers of new computer skills and their achievements:

Robbie Barrat, a self-taught programmer, uses artificial intelligence to incorporate hip-hop and fashion design elements into artistic creation.

I trained a neural network that can write lyrics.

When ROBBIE BARRAT was at a rural high school in West Virginia, he began buying old computers from a local recycling center, taking them apart and putting them back together, then learning to code on his family's farm.

After discussing with friends whether computers are creative or not, he took AI and related courses in high school. To prove that computers are also creative, Barrat trained rap neural networks on Kanye West's lyrics. Barrat's buddies like the AI system very much, but the result shocked some adults. "the teacher was a little upset because they said the neural network was'a little vulgar."

This somewhat mysterious artificial intelligence system helped Barrat get out of the farm. His grades are not good enough to enter an ideal school to study math or computer science. But the program helped him get an internship in a self-driving program in the heart of Silicon Valley. Since then, he has moved to Stanford and is now working in a biomedical laboratory at Stanford, trying to develop neural networks that can recognize molecules with medicinal potential. However, it is still his passion to cultivate neural networks to make art.

Now, in her spare time, Barrat uses video clips and photos from fashion shows to create AI-generated images of models in new clothes. The result was not satisfactory-did you ever think that you would like pants wrapped around your calves, or sweaters with huge bags hanging from one side? However, Barrat is working with designers to make them real clothes. He can't wait to try.

Shaza Mehdi, a computer science freshman at the University of Georgia, trained a neural network to identify plant diseases.

Diagnosis of plant diseases? Just download a program.

Shaza Mehdi's front yard planted a flower called ROSEBUSHES. This kind of flower is beautiful, but it is easy to get sick. One day last year, Mehdi suddenly asked herself: why can't she use her cell phone to diagnose the disease of this flower? Then she met the neural network.

Mehdi does not know how to program, and adults around her can give her encouragement, but not professional knowledge; her school does not offer an introduction to computer science. So Mehdi taught himself the basics of the programming language Python and neural networks through YouTube videos and online tutorials. Now when she recalls what happened during her study, she will excitedly mention her experience of debug.

Mehdi was particularly inspired by a YouTube video from a Stanford researcher. The researchers built a neural network that recognizes skin cancer at a level comparable to that of a dermatologist certified by the medical association.

She found an online tutorial to reproduce neural networks. The first step is to download software that is trained to identify everyday items such as toilets and teapots. The second step is to re-adjust its visual parameters by inputting about 10000 marked images of sick plants. These images are tirelessly collected by Mehdi on the Internet, classified according to different diseases.

At the end of 2017, she finally tested the app, which she named plantMD. Mehdi looked nervously at a sickly vine with light green and brown spots on its leaves. A leaf with a dent suddenly focused on the phone screen. Before the heart could beat a few times, the phrase "grape anthracnose" appeared on the screen. An online search confirmed the diagnosis: this is an obvious case of fungal infection, also known as bird's eye decay.

Dayou Tanahara learned about machine learning online and used it to automate part of his dry cleaning business.

A camera that can check clothes at any time.

Dry cleaning is an arduous task in Japan's aging small cities. Mr Tanahara's family owns eight dry cleaners in Tanagawa, a southern prefecture-level town of about 50,000 people, where it is hard to find good staff. So Tanahara began to think about letting computers increase his labor force.

At first, the 38-year-old tried to modernize his business with better computer systems to record and track orders. But most of his employees have little technical experience and are difficult to adapt.

"they have a bad memory," Tanahara said. So the self-taught programmer began to study how software can automatically check by simply looking at the customer's clothes. He reads about machine learning on the Internet and tries to improve his English and programming skills. In the store, he took 40,000 photos of suits, shirts, skirts and other clothes and used them to train his code.

In July, Tanahara began testing his system in one of his stores. Customers put their clothes on the table and install cameras above their heads. His software looks at the clothes and gives predictions (for example, two shirts and a jacket) and then asks the customer to confirm it on the tablet. At first, employees have to help customers adapt to this approach in the first place. After that, customers can use it alone.

Mr. Tanahara said his workers were initially skeptical of his innovation, but that attitude was changed after they found that the innovation made their jobs easier. He does not intend to use the project as an excuse to fire human employees, but he hopes it will help him expand the store. "I want to open a store with machines and no employees," he said.

Will Roscoe follows his self-driving car in Oakland, California. He wants to prove the feasibility of the concept of automatic public transportation through this car assembled by all kinds of parts.

Miniature self-driving car

In a warehouse in Oakland, California, a small group of nerds gathered to watch Will Roscoe tap his phone with his thumb.

At his feet, a remote-controlled car with a torn plastic skin began to drive around the track marked with yellow and white tape on the scratched concrete floor-Roscoe gave no further instructions. The nearly Frankenstein-style car has a camera and a pile of electronic components, which are zipped to the top. Its big name is donkey cart.

Roscoe is not an artificial intelligence expert, but his creation uses neural network software similar to the software Google's driverless car Waymo relies on to help cars perceive the environment.

As a trained civil engineer, Roscoe was inspired by political frustration to create donkey carts. In 2016, he ran for the board of directors of BART, the Bay area subway system. Roscoe promised to increase passenger capacity by replacing trains with self-driving electric buses, but he only finished third. Building your own small self-driving car seems to be a good way to show voters that the technology is not pure fantasy. "I want to prove that it can work on a small scale," he said.

His timing proved perfect, and hacking RC cars, a group of robot enthusiasts specializing in remote-controlled cars, held its first meeting in nearby Berkeley shortly after. There, he met Adam Conway--Adam and offered to help build the car.

Roscoe is a self-taught programmer who built an autopilot using TensorFlow. He also borrowed some neural network code from RC car party participants. The final product of the Roscoe learns to drive by watching humans manipulate the vehicle during the demonstration operation.

He named his creation Donkey Car because he thought it was very much like a donkey in a way: very safe for children, not classy in traditional aesthetics, and often disobedient to instructions.

Roscoe and Conway put all their software and hardware designs online for others to use. Donkey carts (Donkey Cars) are now held in Hong Kong, Paris and Melbourne, Australia.

In January, at a warehouse in Oakland, nine homemade self-driving cars tried to compete on the track; one of the competing cars was a donkey cart built by three high school students.

These vehicles are also starting to play a role outside the track. Two amateurs near Los Angeles modified them to find and dig rubbish on the beach.

There are a lot of leaves stuck on Oakland,Roscoe 's donkey cart suspension. "often take it to the sidewalk."

"look, I even have the leash ready." Roscoe said.

Related reports:

Https://www.wired.com/story/diy-tinkerers-artificial-intelligence-smart-tech/

Welcome to subscribe "Shulou Technology Information " to get latest news, interesting things and hot topics in the IT industry, and controls the hottest and latest Internet news, technology news and IT industry trends.