摘要
在人工智能高速发展的时代,对机器学习领域的探索占据重要的地位,而机器学习本质上源于对海量数据的分析与学习,这就离不开统计学中模型的建立与推断。贝叶斯方法作为统计学中主要且成熟的建模方法,在充分学习样本信息的前提下引入参数的先验信息,容纳了参数的不确定性,使模型推断更加合理。在贝叶斯框架下的非参数方法进一步扩大了这种不确定性,将参数的先验空间推广到分布空间,用随机过程来进行表示,此时的先验空间是无限维的。贝叶斯非参数建模方法以其巨大的灵活性和稳健性得到了广泛的关注,随着人工智能的迅速发展,研究人员在机器学习领域对贝叶斯非参数方法展开了深入的研究并取得了许多优异的成果。本篇论文探究了贝叶斯非参数的部分基础理论,并对其在大数据背景下的实际应用进行了研究与展望。
In the era of rapid development of artificial intelligence, the exploration of the field of machine learning occupies an important position, and machine learning essentially stems from the analysis and learning of big data, which cannot be separated from the establishment and inference of models in statistics. Bayesian methods, as the main and well-established modelling methods in statistics, introduce a priori information about the parameters with sufficient learning of sample information, accommodating the uncertainty of the parameters and making model inference more reasonable. Nonparametric methods in the Bayesian framework further extend this uncertainty by extending the prior space of parameters to the distribution space, which is represented by a stochastic process, at which point the prior space is infinitely dimensional. Bayesian nonparametric modelling methods have received widespread attention for their great flexibility and robustness, and with the rapid development of artificial intelligence, researchers have conducted in-depth research on Bayesian nonparametric methods in the field of machine learning and achieved many excellent results. This paper explores some of the underlying theory of Bayesian nonparametric and investigates and prospects for its practical application in the context of big data.
出处
《统计学与应用》
2023年第2期283-292,共10页
Statistical and Application
