When most of us think about the fundamental language of computational devices we think of ones and zeros; discrete representations of on and off. While this is overwhelmingly the case in modern computational devices, it was not always thus. Pre-1960's technology, such as that used in the second world war, often performed calculations using analogue computation. In an analogue computer, calculations are performed by using an "effectively" continuously-changable physical quantity such as electric potential. This has some significant advantages as well as many disadvantages, for a detailed discussion see: Analog Versus Digital: Extrapolating from Electronics to Neurobiology; Sarpeshkar. (A highly recommended read!)
Perhaps the most interesting aspect of analogue computing is how neurological systems use it in conjunction with digital signals (again see Sarpeshkar for details). There are two obvious aspects of analogue information processing in neurobiology, temporal encoding and the integration of passive membrane signals by the neuronal cell body. Temporal encoding works by coupling a neurons firing rate to a particular brain-wave's phase (by brain-waves I mean the background oscillations in brain activity, not ah ha! moments).
Perhaps the most crucial aspect of neural computation however is the structure of the networks themselves. Siegelmann and Sontag provide a theoretical model of analogue computing within neural networks.
Analogue computing is an understudied and (in my opinion) a fascinating field of research.
No comments:
Post a Comment