AI Technology Revolution

The Barron Weekly pointed out that artificial intelligence technology is a hot topic in recent days, but this wave of revolution has only just begun. In the process of developing AI, it needs to use huge computing power. For semiconductor companies such as Micron Technology and Xilinx, it is very good news. They may become future chip stars.

As early as October 2015, Barron Weekly had a cover story on the theme of AI. At that time, it was titled "Watch Out Intel, Here Comes Facebook". In the case of Nvidia, which was once mentioned, the share price has soared by 770%. However, Micron may become a new star in the next generation chip industry under the support of AI development. Because the development process requires a lot of memory chips, the stock price of the company has grown by 270% so far.

Nvidia share price chart for five years

At present, AI is still in the early stages of development and there may be more powerful returns in the future.

With the advent of network and cloud computing, network giants can aggregate large amounts of data and perform calculations in their data centers. Computer scientists have created a new machine learning model called deep learning. This model does not require scientists to establish explicit calculation rules.

The following figure shows an example of deep learning in which a computer is taught to recognize pictures of cats, millions of pictures of cats are entered into computers, and many examples of non-cats are also entered, such as photographs of dogs. The computer will detect In image pixels, there are many basic shape patterns.

One of the deep learning paradigms, identifying cats and dogs

Then it will find these very basic shapes, how to gather to form related (pointed ears) and irrelevant (soft ears) recognizable pattern features, which will become a new model of the computer, and later in identifying whether the picture is a cat It will apply this model.

This kind of deep learning is a powerful new paradigm: Humans only need to set a basic goal, such as classifying pictures, and then let the computer find out the solutions. The series of steps the computer goes through is called “network”. Different networks are good at different things. For example, so-called reinforcement learning uses very simple messages. It only requires a scenario and a set of rules. They can find the best mode of action.

At the end of 2017, Google used this method to create a self-learning computer Go AI model AlphaGo Zero. This model uses neural network technology. It learns various moves and discovers new strategies by playing against itself. He is The human intelligence is not needed in the learning process and it eventually defeats top human Go chess players.

The important role that technology companies play in the development of AI is to provide the necessary hardware support or new machine equipment. For example, Nvidia's GPU chips are more efficient than Intel's microprocessors. Micron's demand for memory chips has soared, and it is expected that In 2021, chips that support deep learning were introduced.

Deep learning will become more common in computing. Over time, deep learning will take over the work of humans writing programs. However, due to the development of chips and software, there is still a long way to go to achieve this goal.

2016 GoodChinaBrand | ICP: 12011751 | China Exports