Technological Educational Institute of Crete, School of Applied Science, Estavromenos, GR 71004 Iraklio, Crete, Greece.
Neural Comput. 2013 Sep;25(9):2265-302. doi: 10.1162/NECO_a_00476. Epub 2013 May 10.
In this work, the Shannon information transfer rate due to the transmission of a linear combination of the firing rates of a number of afferent neurons is examined. The transmission of this linear combination (transfer statistic) takes place through a stochastic firing process, while a rate code is assumed. Constraints are imposed on the transmission process by the requirement that the coefficient of variation for the transfer statistic is small and by the relative variance of the individual terms in the calculation of the statistic. In the regime of no noise or signal correlations among the input neurons, simulations suggest that information transfer for fixed overall input is favored when there are few high-firing neurons, as opposed to more lower-firing neurons. Signal correlations among low-firing neurons can result in aggregates of high firing rates, improving in this way information transfer and calculational robustness. Under reasonable rate code assumptions, information transfer rates obtained are of the order 3 to 10 bit/sec.
在这项工作中，我们研究了由于传递一些传入神经元的发放率的线性组合而导致的香农信息传递率。这种线性组合的传递（传递统计量）是通过随机发射过程进行的，同时假设存在一个速率码。传输过程受到约束，要求传递统计量的变异系数小，并且要求在计算统计量时各个项的相对方差小。在输入神经元之间没有噪声或信号相关性的情况下，模拟表明，当存在少量高发射神经元而不是更多低发射神经元时，固定的总输入的信息传递更有利。低发射神经元之间的信号相关性可能导致高发射率的聚集，从而以这种方式改善信息传递和计算稳健性。在合理的速率码假设下，获得的信息传递速率约为 3 到 10 比特/秒。
