ABSTRACT
The aim of neural coding research is to identify the principles by which neural circuits encode, communicate, and process information. The idea that emerged and dominated in the rst half century of work, starting with Adrian, was that the neural code is a straightforward, low bandwidth “rate code,” where the basic element of the code is the spike count-the number of spikes emitted within an “encoding window” of order 1 s (Adrian and Zotterman, 1926). It has long been recognized, theoretically, that much more powerful neural codes are possible, if neurons are able to emit temporal patterns of spikes whose timing is reliable on a shorter timescale (MacKay and McCulloch, 1952). These might be either temporal patterns of spikes emitted by a single neuron, or temporal patterns coordinated across a population of neurons. There has, consequently, long been interest in such “temporal codes” (Perkel and Bullock, 1968). However, there has been skepticism whether temporal codes are sufciently robust in the face of noise to be neurobiologically viable (Shadlen and Newsome, 1994; London et al., 2010). This reects a fundamental and fascinating debate in neuroscience-is neural computation based on noisy devices, constrained to communicate information at low bandwidth or, alternatively, on efcient, high bandwidth devices that push the physical limits of spike train coding?
