The goal of this chapter is to consider some of the fundamental implications of the event sequence nature of neural activity. We approach this, primarily, from the point of view of function: we will look at some simple, idealized vignettes that highlight the implications of the event-sequence nature of neural activity for representing and processing information. While the issues that these vignettes raise are all related to spike timing, they are also logically distinct from each other, and therefore deserve separate consideration.
The starting point is the well-known fact that neurons communicate with each other via a sequence of action potentials, a profound observation that has been with us for the better part of a century. The reason that the observation is so profound is that it has implications for how the brain operates at the most fundamental level. This is because action potentials ("spikes") are stereotyped -- for a given neuron, each spike is largely identical to every other one. Thus, in order for a neuron to transmit information, it must vary the timing of the sequence of spikes that it emits. This, of course, is the motivation for the collection of topics discussed in this book:understanding how spike sequences represent information requires understanding the mechanisms for encoding signals into temporal sequences of spikes and the mechanisms for decoding these sequences, as well as having appropriate analytical tools for these investigations.
However, the event-sequence nature of neural activity has a number of distinct implications beyond spike timing per se. Perhaps the most basic is that it forces us to carefully consider the kind of mathematical entity used to represent spike trains. This in turn, has implications for how we formulate models for the evolution of neural activity over time, and theways in which we analyze how neural activity can represent information.
The first vignette moves from these abstract considerations to a very concrete aspect of the event sequences: they can contain a positive number of events, or no events, but they cannot contain a negative number of events. This elementary asymmetry has theoretical implications for how increments and decrements can be detected: we illustrate the basic idea, and then presentsome data from a model system (the subdivision of ON and OFF signals in the retina) that suggest that these ideas are in fact relevant to real neural circuits.The last two vignettes address how a discrete sequence of events can represent a continuous quantity. In broad terms, this can happen two ways: in time and in space. To focus on the temporal strategy, we consider how a single neuron can represent a continuous quantity by varying its spike rate over an extended period of time. To focus on the spatial strategy, we consider how a population of neurons can represent a continuous quantity via a distributed pattern of activity at a single instant. Although we look at highly simplified scenarios in both cases and take an elementary viewpoint, these vignettes nevertheless serve to illustrate the impact of the event-like nature of neural activity on the roles of noise, variability, and neuronal diversity.
Finally, we emphasize that our goal is to describe the range of possibilities afforded by event sequences, not to predict how the brain should operate. While we do not doubt that evolution has pushed the brain towards some kind of optimality, the standards of optimality are very complex, and include many biologic constraints beyond the ones considered here:constraints of chemistry and physics, constraints of the genetic code, constraints of development,the need to adapt successfully to an environment that changes over short and long timescales, etc. Thus, the simple vignettes we consider make no attempt to deduce optimal strategies, but serve only to illustrate how the event-like nature of neural activity shapes the repertoire of strategies that the brain can use, and, consequently, the ways in which we need to approach experimental data.