Beijing time on September 24th news, IBM has always had a dream, the development of a computer, it can make decisions like a man, with intelligence. Recently, IBM research has made a major breakthrough, it is closer to the goal.
IBM has developed a chip named TrueNorth, can simulate the human brain operation. In order to prove that it is very fast, very low energy consumption, IBM tested.TrueNorth has deep learning ability of chip, like a human brain, it can also be used for correlation analysis to determine the possibilities. Although some other computers can also have the same ability. But they are a lot of high energy consumption.
Because TrueNorth can study, in the future, we can
IBM previously demonstrated the chip, then it will be loaded into a chip called NS16e computer, the operation principle of the computer and the human brain is similar to the use of neural networks, NS16e can identify the image, voice, mode.
Human brain is extremely complex, with 100 billion neurons, neurons through the nerve node
IBM said, TrueNorth processor can be classified to the picture data, the speed of 1200-2600 per second frame, energy consumption of 25-275 MW. The processor can also recognize the picture mode, these pictures is a 50-100 camera shooting speed of 24 frames per second. The future, IBM functionality can be embedded into the intelligent mobile phone, a few days without charge can be used.
TrueNorth, the energy consumption is low, the existing installation is the traditional chip, such as GPU, CPU and FPGA, they can recognize the image and voice of.Facebook, Google, Microsoft and Baidu are learning techniques developed in depth, through analysis and pictures, voice relevant answers, deep learning network can automatically learn GPU general drive network and the energy consumption is very high, more than 150 watts.
IBM developed for the TrueNorth algorithm and deep learning mode, which includes pattern recognition, can contact the past data and current data together. According to the different patterns of deep learning IBM development, the corresponding algorithm, however, the chip is still compatible with the existing system, such as MatConvNet., in other words, developers can use MatConvNet IBM chip the development of learning model, TrueNorth can handle tasks in the background, the developer does not need to directly face the TrueNorth.
This process and early game development is somewhat similar to the beginning, the game developers are not directly into the GPU, they don't know how to use the function of GPU. Until recently, Vulkan replaced OpenGL API, developers can directly use the GPU function.
Deep learning is very strong, the driverless car is clear, powerful computer navigation, vehicle identification signal, lane and other objects, the safe driving.TrueNorth chip makes every neuron low-level processing, and then the results will be together, eventually identifying images or sounds. In fact, Nvidia, using Intel chips the technology is the same, just a lot of high power consumption.
TrueNorth is still in the early stages of development, in accordance with the IBM plan, it is ready to develop a computer and the human brain, the problem is that the preparation algorithm and application of some challenges.
In 2004, IBM began to study