The world’s first blockbuster | photon neural network is born: 1960 times faster than traditional methods of Sohu technology selected from the arXiv machine of the heart: Wu Pan before the compiler in Alexander N. Tait from the Princeton University and other scientists in the arXiv published an article entitled "neuromorphic Silicon Photonics (Neuromorphic Silicon Photonics)" the introduction the world’s first "photon neural network (Photonic Neural Network)". Click to read the original text to download this paper. The artificial intelligence technology represented by machine learning neural network is changing many aspects of our life, and the demand for data processing ability to support these technologies is becoming stronger and stronger. The development of neural network chip (neuromorphic chip) is expected to significantly improve the performance of neural network. The day before, from the Princeton University Alexander N. Tait and other scientists at arXiv published an article entitled "neuromorphic Silicon Photonics (Neuromorphic Silicon Photonics)" the paper, introduced the world’s first "photon neural network (Photonic Neural Network)". Compared with the current mainstream electronic processing methods, we can achieve faster and higher bandwidth. According to the report of the Technology Review MIT, the core of the problem is to develop a device for each node to have the same response characteristics as a neuron. These nodes are in the form of a miniature annular waveguide (tiny circular waveguides), which is etched in a silicon substrate, where light can be recycled. When the light is released and the output of a laser modulated to work at the threshold, a slight change in the input light will have a huge impact on the output of the laser. The key is that each node in the system works at a certain wavelength of light – a technique known as multiplexing division (wave). Before the light from each node is fed into the laser, the sum of the total power is detected. The laser output is then fed back to the node to create a feedback loop with nonlinear characteristics. So to what extent does this kind of nonlinearity simulate the neural behavior. The results show that the output is mathematically equivalent to the output of a (continuous-time recurrent neural network). In order to verify the concept, the researchers developed a 49 node silicon photonic network. The experiment results show that the experimental results show that the proposed method is faster than the traditional method in an experimental simulation of differential systems 19n相关的主题文章: