自然界信号的1/f特征和感觉神经系统响应特性

1/f Characteristic in Natural Signals and Sensory Neural Response Properties

众多实验证据表明,大脑感觉神经系统对自然信号的处理比对人工信号有着更高的效率,这可以理解为是自然进化的结果.那么大脑神经系统究竟是＂适应＂了自然信号的哪种统计特性？或者说神经元或神经网络信息处理的数学函数究竟对应什么样的自然信号特性,使得其可以发挥最大的响应性能？搞清楚这个问题对于理解大脑工作机制和发展类脑人工智能具有重要意义.与此相应,本研究组了解到自然界信号普遍存在着一种统计特性—1/f特征,即信号的功率谱能量密度P（f）随频率f增加而依幂律下降,其关系可以被描述为P（f）~1/f~β,（β~=1）.例如,自然界的花草树木景色、海水的流速、心脏跳动、音乐、语言、大脑脑电信号等均呈现出1/f特征.虽然至今1/f特征产生的物理机制仍然是个迷,但很显然,研究清楚大脑神经系统如何响应1/f信号特征,以及理解大脑脑电信号形成1/f特征的神经过程,对于理解大脑对自然界信号的神经信息处理和信号表征机制至关重要.本文将回顾这一研究方向的相关实验和理论证据.

Abundant experimental evidence has shown that sensory neurons process natural signals more efficiently than artificial signals. Although this is understandable based on evolution, it remains unclear which types of intrinsic signal properties of natural signals are used by the brain to ＂adapt＂ and make processing efficient. What type of signal parameters stimulate neuronal or network functions to their maximal performance？ The answer might be critical for understanding the operating principle of brains and important for the development of artificial intelligence. Correspondingly, it is well-known that the 1/f characteristic（or long-term correlation） is a common feature in the power spectra of natural signals. The power spectral density P（f） of a natural signal（for example, the temporal luminance variation in natural scenes, velocity of ocean waves, loudness of natural sounds, variance of heart-rate, spontaneous neuronal activity, etc.） typically decreases in power as frequency increases, showing a mathematical relationship as P（f）~1/f~β（where β~=1）. Although the physical mechanism of the generation process of the 1/f characteristic remains unclear, it is important to investigate how the nervous system produces a response to signals with 1/f characteristics and how the brain electronic activity formulates 1/f-type signals. This information may be fundamentally relevant to understanding the neural mechanism of information processing and signal representation. This paper will review the theoretical and experimental progress related to this issue.