统计认知理论及应用研究

Statistical Cognition Theory and Its Applications:An Overview

长期以来大家认为人类认知尽管可以看成是非确定的推理计算过程,但它的知识表达、模型结构、及计算方法和概率统计理论在本质上是不同的,因此认知科学和概率统计方法存在巨大的鸿沟,过去两者基本上独立发展。近年来随着Bayesian概率统计模型研究的一系列突破性工作和认知过程本质的不断被发现和挖掘,两者的相关性和互补性逐渐突显出来。许多研究者认为认知是近似遵循概率统计推理原则的,一些研究工作显示两者的结合有可能对人工智能发展产生深远的影响。本文对当前统计认知理论及应用研究的现状进行系统的梳理,并结合自身的研究对它今后的发展提出自己的看法。

The notion is popularly accepted that even though human cognition can be considered as an uncertain reasoning process, it is still essentially different from probabilistic theory in knowledge presentation, model structures, and computational mechanism, therefore there is a large gap between cognition science and probabilistic statistics, and these two theories are developed independently. Recently, remarkable progress in Bayesian statistical models and rediscovery of cognitive process, make their correlation and cooperation attract more and more attentions. Many researchers think that cognition approximately follows the rational principles of statistical inference, and some research results had shown their combinations would produce an important effect on artificial intelligence research. This paper will give a brief overview of statistical cognition as it stands today, and discuss some future research.