What is the neural code?
The interpretation of how spiking activity conveys information has always been a central problem in neuroscience. That is, what information is the brain “looking at” when its decoding a spiking sequence? The most commonly used interpretation – or neural code – is the rate code, which states that the brain is understanding the signal through a time-averaged value over a certain window. On the opposite side of the spectrum is the spike code, which believes that precise spike times carry information itself. Other intermediate ones, such as correlation-based code, have been proposed as well. Correlation-based code believes that while information is coded by firing rate, the correlations between different populations of cells indicate whether they speak of the same object or not (i.e. the binding problem). While there are many candidates for the neural code, most of the experimental evidence is weak. The only candidate with reliable experimental findings is the rate code, and hence this chapter will focus entirely on the rate code – specifically, on its criticisms and rebuttal of those criticisms.
The greatest opposition of the rate code is the efficiency hypothesis. It is often observed that the brain has irregular spiking activities. For rate codes, such irregularities are often interpreted as noise (while spike code would argue that these spikes are informative), and for long time windows, a neuron firing regularly would lead to a more accurate representation of the targeted firing rate. However, experiments have shown that visual processing is fast, and therefore the time window for time averaging must not be long. Under these circumstances, it can be shown that the regular and irregular firing have similar estimation errors. In fact, it is easy to create a random network that fires regularly. To create an irregularly firing network, the input to each neuron must be near threshold, and the fluctuations must be large. This condition can be met when the network is balanced. Van Vreeswijk and Sompolinsky showed that a balanced network has the advantage of being able to react swiftly to changes in input, and enhances input modes of interest. From this perspective, then, the irregularity almost seems like a mere byproduct.
For the reasons above, the author argued in favor of the rate code, at least in brain areas such as the primary visual cortex. This doesn’t necessarily apply to other brain areas, for example the primary auditory cortex. Being able to understand the neural code within different parts of the brain would give insight to both theoretical and experimental work in neuroscience.
Summary written by: Belle Liu
Original article: Van Vreeswijk, C. (2006). What is the Neural Code? In J. L. van Hemmen & T. J. Senjnowski (Eds.), 23 Problems in Systems Neuroscience (pp. 143-159). Oxford University Press, Inc.
The greatest opposition of the rate code is the efficiency hypothesis. It is often observed that the brain has irregular spiking activities. For rate codes, such irregularities are often interpreted as noise (while spike code would argue that these spikes are informative), and for long time windows, a neuron firing regularly would lead to a more accurate representation of the targeted firing rate. However, experiments have shown that visual processing is fast, and therefore the time window for time averaging must not be long. Under these circumstances, it can be shown that the regular and irregular firing have similar estimation errors. In fact, it is easy to create a random network that fires regularly. To create an irregularly firing network, the input to each neuron must be near threshold, and the fluctuations must be large. This condition can be met when the network is balanced. Van Vreeswijk and Sompolinsky showed that a balanced network has the advantage of being able to react swiftly to changes in input, and enhances input modes of interest. From this perspective, then, the irregularity almost seems like a mere byproduct.
For the reasons above, the author argued in favor of the rate code, at least in brain areas such as the primary visual cortex. This doesn’t necessarily apply to other brain areas, for example the primary auditory cortex. Being able to understand the neural code within different parts of the brain would give insight to both theoretical and experimental work in neuroscience.
Summary written by: Belle Liu
Original article: Van Vreeswijk, C. (2006). What is the Neural Code? In J. L. van Hemmen & T. J. Senjnowski (Eds.), 23 Problems in Systems Neuroscience (pp. 143-159). Oxford University Press, Inc.
留言
張貼留言