Decryption deep learning
　　 1.1. The development of artificial intelligence has been fluctuating along with the progress of artificial neural network research.
The history of the entire artificial intelligence development has almost always fluctuated along with the progress of artificial neural network research. In the near future, the deep learning of artificial intelligence has a new round of enthusiasm. The “depth†in its name refers to the number of layers of artificial neural networks. Deep learning is essentially a machine learning algorithm based on multi-layer artificial neural networks.
1.2. What is artificial neural network?
The information activity of the human brain has three distinct characteristics compared to current computers:
First, massive parallelism and fault tolerance. There are about 100 billion neurons in the human brain. There are about one trillion synaptic connections between neurons, which form a labyrinthine network connection. A large number of neuron information activities are carried out simultaneously, instead of the current computer according to the instructions. Execution. In addition, this massive parallelism of the human brain also makes it extremely fault-tolerant. A broken transistor can destroy a microprocessor, but the brain's neurons die all the time.
Second, the information processing and storage unit are combined. At present, the computer generally adopts the von Loyman architecture, and the memory and the processor are separated, and the data is transmitted through the bus. As the amount of processed data grows massively, the limited data transfer rate of the bus is called the “von Neumann bottleneckâ€, which seriously affects the computational efficiency and power consumption of the computer. The human brain information processing and storage unit are combined and have Very low power consumption (about 20W or so).
Third, self-organizing self-learning functions. The brain learns and changes while interacting with the outside world, rather than the fixed path and branching of the computer that follows the preset algorithm.
Based on the above points, people have been trying to imitate the information activity mechanism of human brain neurons to design algorithms: signals enter the neuron cells through synapses, and the nerve cells use a way to make all the signals coming in from the dendrites. Plus, if the sum of all the signals exceeds a certain threshold, the neuronal cells will be excited into an excited state, and an electrical signal will be sent out through the axons to other nerve cells. If the sum of the signals does not reach the threshold, the nerve cells will not be excited and will not transmit signals.
A simple artificial neuron mathematical model is to let each signal input to the neuron be weighted and summed. If it is added, if it exceeds the set threshold, it will output "1", and if it is not, it will output "0". Such a combination of several simplest neuron inputs and outputs constitutes a complex artificial neural network.
Industrial Diesel Generator,Industrial Generators,Commercial Diesel Generators,Industrial Type Diesel Generator
Jiangsu Vantek Power Machinery Co., Ltd , https://www.vantekpower.com