Part of the content of the article is referenced from captainbed.net

[TOC]

Overview of Neural networks

  • Previous AI (clustering, expert systems, inductive logic) was fake intelligence, more statistical intelligence than ARTIFICIAL intelligence. It’s just a big, complicated program, and we still know how it works inside.

  • A neural network, on the other hand, is a black box, and we have no idea how it works.

Theoretical basis of neural network

Neural network is inspired by the human brain nerve cells and constructed, each node receives a number of weights input, if certain conditions, output, otherwise not output.

Where x represents the input training set and y represents the output result.

Neural network related concepts

  • The more complex the network, the more powerful the neural network is so we need deep neural networks. Depth is the number of layers, and the more layers you have the more complex the neural network is.

  • The process of training a deep neural network is called deep learning and once the network is built, all we have to do is keep feeding the training data into the neural network, and then it will keep changing and keep learning, and we don’t know how it learns.

  • Each data input to the neural network is called a feature, such as an image recognition neural network with an image of 64*64*3 pixels (64*64 is the resolution and 3 is the three primary colors). The 12288-dimensional picture is also called the feature vector with 12288 features.

Summary of Basic concepts

Some basic concepts commonly used in ARTIFICIAL intelligence are as follows:

sample

The samples are the examples used to train the model. There are labeled samples and unlabeled samples.

The label

Tagging is what we’re trying to predict, for example, whether a picture of an animal is a cat. “Cat” is the label; Or predict the price of the house, and the “price” is the tag. The label can be determined.

⭐ ️ characteristics

Features are input variables, == can be understood as “factors” ==, there may be millions of features in complex machine learning projects. In spam detection, the characteristics might include: the word in the message, the sender’s address, the time it was sent, the frequency it was sent… (It’s not hard to imagine that all of these factors could influence the outcome of determining whether the current message is spam.)

Features that occur more frequently and are more associated with tags are high-value features, and we want to avoid features that occur less frequently, such as the ID of an email.

⭐ ️ model

The model defines the relationship between features and tags. For example, a spam detection model might associate the presence of certain advertising words in a message with “spam.” Two phases of the model lifecycle:

  1. Training: create the model and let the model learn, provide the model with labeled samples, and let the model learn the relationship between features and labels.
  2. Inference: Apply the trained model to the unlabeled sample and let the model make the actual prediction.

That is to learn first, learn to use.