Information-Theoretical Aspects of Embodied Artificial Intelligence
Information-Theoretical Aspects of Embodied Artificial Intelligence
复制标题
实体人工智能的信息理论方面
DOI：
--
复制
发表时间：
2003
期刊：
影响因子：
--
通讯作者：
Teresa K. Pegors
中科院分区：
文献类型：
--
作者：
O. Sporns;Teresa K. Pegors
体现人工智能是自主智能系统设计的一种新方法。这一章是关于这种系统设计的新原则，它深深植根于体现的概念。具体化行为对感觉输入的性质和统计有因果影响，而感觉输入又可以驱动神经和认知过程。通过使用信息论的方法，特别是对感觉数据流的熵、互信息和复杂性的测量，可以捕获感觉输入的统计信息。概述了几种这样的方法，并讨论了它们在具体化人工智能系统中的应用。
Embodied AI is a new approach to the design of autonomous intelligent systems. This chapter is about a new principle for the design of such systems that is deeply rooted in the notion of embodiment. Embodied action has causal effects on the nature and statistics of sensory inputs, which can in turn drive neural and cognitive processes. The statistics of sensory inputs can be captured by using methods from information theory, specifically measures of entropy, mutual information and complexity, on sensory data streams. Several such methods are outlined and their application to embodied AI systems is discussed.