The last two decades of Studies in diffusion models: A bibliometric analysis

Given the importance of diffusion models in daily life and economic growth, developing efficient and accurate diffusion models is an important task. For example, the combination of diffusion models with computer vision is a specific area of studying diffusion models to meet social needs. Many activities within this field have benefited from recent technological advances. The diffusion model is inspired by the theory of nonequilibrium thermodynamics. First, according to the Markov chain definition, random noise will be added slowly in, and then the model has to learn the reverse diffusion process to construct the desired data sample from the noise. The research of diffusion model is the result of the multi-disciplinary integration, and it is also the practical development of the application scope, practicability and efficiency of text generation picture technology.^[1] The application of diffusion model can promote the development of a number of technologies in this field, including computer vision, speech generation, bioinformatics, natural language processing, NLP, waveform signal processing, multi-modal modeling, molecular map modeling, time series modeling, antagonistic purification and other fields. The research content of the diffusion model is more cutting-edge, and its wide application is more in line with the needs of the future society. However, as a cutting-edge technology, the research of diffusion model is not perfect, and there are still problems such as slow sampling speed, poor maximum likelihood, weak data generalization ability, single data processing type, and model dimension reduction. Because this review comprehensively analyzes the prints and publications derived from web of science, based on the author, keywords, age and nationality, summarized the important trends and valuable insights. ^[2]