Title :
Non-information-maximizing neural coding
Author_Institution :
Comput. & Software Syst., Univ. of Washington, Bothell, WA, USA
Abstract :
Information theoretic techniques are often used to investigate neural coding. Results - in terms of bits per second or bits per spike - have been used as evidence to support temporal or rate coding, spike timing precision, etc. Despite its use this way, information theory does not tell one what the neural code (or any code) is. In artificial systems, codes are often purposefully made sub-optimal from a pure information density point of view. This work tests the feasibility of a neural code containing error correction characteristics which uses greater spike timing precision than might be necessary to simply transmit a given amount of information. A model of the recognized prototype of an inhibitory synapse shows that, even compared to small input imprecision and in the presence of robust dynamical behaviors, high timing precision can enable error correction.
Keywords :
information theory; neural nets; error correction characteristics; information density; information theoretic techniques; inhibitory synapse; noninformation maximizing neural coding; rate coding; robust dynamical behaviors; small input imprecision; spike timing precision; spiking neurons; temporal coding; Channel capacity; Error correction; Error correction codes; Information theory; Neurons; Prototypes; Robustness; Software systems; Testing; Timing;
Conference_Titel :
Neural Networks, 2003. Proceedings of the International Joint Conference on
Print_ISBN :
0-7803-7898-9
DOI :
10.1109/IJCNN.2003.1223999