TABLE 1.

Comparison table of different ML subtypes and DL structures.

ML subtypes	Key features	Major use
Supervised learning	Learns from input‐output pairs to predict or classify new input data	Classification and regression
Unsupervised learning	Input data without corresponding labels, learns to discover patterns in the data on its own	Clustering, dimensionality reduction, and anomaly detection
Reinforcement learning	Make optimal decisions by interacting with an environment	Robotics control and autonomous driving
Self‐supervised learning	Uses the characteristics of the data itself for supervision	Image, video, and speech recognition
Weakly supervised learning	Incomplete/inaccurate/inexact data labels	Image and speech recognition
Active learning	Representative samples are selected for annotation to better train the model	Image and speech recognition

DL structures	Key features	Major use
CNN	Convolutional layers and pooling layers, which can effectively extract features from images	Image and speech recognition
RNN	Recurrent layer, which can process data with front and back correlation	Natural language processing, speech recognition
GAN	Two neural networks, a generator and a discriminator, which generate realistic samples through adversarial training	Image generation, video generation
Transformer	Self‐attention mechanism and multi‐head attention mechanism, which can process long text sequences	Machine translation, text generation, question answering

Abbreviations: CNN, convolutional neural network; DL, deep learning; GAN, generative adversarial network; ML, machine learning; RNN, recurrent neural network.