Speaking of the hottest thing recently, it must be AI, and everyone must be very curious about how AI works.
Given that the current models in various fields are based on deep learning, let's explain how deep learning AI works in the best understandable language.
AI is artificial intelligence, let's not talk about complex AI definitions, just understand that AI is a simulation of the brain.
When people process information, they often collect information around them first, take Go as an example, people will observe the distribution of black and white sub-on on the chessboard, and then the brain processes this information, and then chooses a place to play.
Here, the distribution of pieces on the board is the input, and what action you take is the output.
How does your brain process this data? Let me give you a grand introduction to neurons.
The input is calculated by the neuron to produce the output
The human brain contains hundreds of different types of neurons totaling nearly 100 billion neurons, and neural signals are received, transmitted and processed through neural circuits formed by synaptic connections between neurons.
The thickness of each neuron is different, and the thickness is different, which determines which way the signal is more likely. If you exercise a skill frequently, you must make a certain neural pathway stronger, so that you will be stronger at this skill.
So how does AI simulate the human brain? Here we are going to talk about a little boring junior high school knowledge.
Let's take a closer look at this input-calculation-output process, like the equations we learned in junior high school? If y is defined as the output and x as the input, then it can be simply expressed as y=wx. The w here is actually the weight, that is, the thickness of our neurons. Of course, W alone is not enough, for a more accurate expression, we need to offset term B. I will not delve into the role of B here, in short, to cover more possibilities.
So we have y=wx+b, which is a node of AI, and when many layers of nodes are stacked, you get an AI.
When we need to train an AI, we need to collect x and the corresponding y, give the data to the AI model, and then let it adjust the parameters w and b. When the training is complete, you enter the new x into the model and you get y.
So in essence, the working principle of graph AI and language AI based on deep learning at this stage is imitating the human brain. The next time someone asks you how AI works, you can tell them that it's the same as your head.