R EVIEW 0166-2236/99/$ – see front matter © 1999 Elsevier Science. All rights reserved. PII: S0166-2236(98)01376-9 TINS Vol. 22, No. 7, 1999 303 C OGNITIVE NEUROSCIENCE hinges on the doctrine that the brain represents the world in patterns of neuronal activity. Recently, this ‘representationalist’ credo was forcefully restated by Blakemore and Movshon 1 . ‘The task of sensory systems is to provide a faithful representation of biologically relevant events in the external environment… These representations are [rich] because they contain representations of objects, states, and events that are abstracted from the primitive sensory signals; they are [simple] because they represent the distillation of the vast quantities of raw measurement information offered to the central nervous system by each sensory surface.’ As early as four decades ago Lettvin et al. 2 anticipated this credo in the classical paper ‘What the frog’s eye tells the frog’s brain’, which phrased the neuronal sensitivities of amphibian tectum neurons in terms of biological, or ecological, relevance of the represented stimuli. The advent of the system-theoretic approach in the 1960s with its stimuli motivated by system identifi- cation and signal-filtering considerations, however, pushed the ecological relevance of the stimuli aside as a poorly defined idea. Needless to say, the system- theoretic approach has proven to be of great value in deciphering many important aspects of early sensory processing, albeit at the price of missing the key question: how do sensory systems function when con- fronted with stimuli that exist in their ecological niches? Probably the most profound difference between con- ventional impoverished laboratory stimuli and eco- logically relevant stimuli is that the parameters of the former stimuli are kept constant for extended periods of time (1 s), while the latter stimuli are in perma- nent flux caused by changes either in the environment or the observer. Recent experiments involving the stimu- lation of neurons in insect, amphibian and mam- malian sensory systems on ecologically relevant time scales revealed not only that response properties were predictable from responses to conventional constant stimuli, but also some exciting hitherto overlooked properties, which, thus, highlighted the importance of the temporal factor in sensory signaling. This article gives a cursory update on the advances in studies of sensory representations that vary on eco- logically relevant time scales (reviewed previously by Bialek and Rieke 3 ). These advances have been stimu- lated by novel information-theoretic approaches that can be used to gauge the ‘richness’ and ‘simplicity’ of neuronal representations of rapidly varying stimuli 4–7 , and by accumulation of neurophysiological data from a wide variety of species. From spikes to representations Before embarking on a survey of recent developments in mapping neuronal representations of ‘eco-relevant’ stimuli, an explanation of the basic assumptions em- ployed in these studies is necessary. The information- theoretic approaches referred to in the previous para- graph aim to provide a rigorous evaluation of the statistical relationship between a stimulus set and neur- onal response. One might argue that the conventional mapping of tuning curves and the measuring of neur- onal sensitivities by means of various selectivity indices are directed towards the same goal. Indeed, these meas- ures have proved to be of great value when neurons exhibit smooth tuning curves with a single maximum (preferred stimulus values) and the response-probability distributions can be satisfactorily characterized by the first two moments (see also Ref. 8). No obvious exten- sions of these measures appear to be available for time- varying stimuli. More importantly, these measures are descriptions of a neuronal signal as a function of stimu- lus. The idea of neuronal ‘representation’ of a stimulus, however, invokes the inverse relationship: that of stimu- lus as a function of neuronal response. Indeed, from the organism’s perspective, only the latter relationship is meaningful as, unlike an experimenter, who is pre- occupied with describing neuronal responses to stimuli drawn from a domain of some sensory feature space, an organism is routinely engaged in the inverse task: inference of stimulus value from a neuronal response 9 . The commonplace measure that comes closest to cap- turing this type of inference is the discriminability analy- sis [that is, receiver operating characteristics (ROC)] derived from the signal-detection theory (see Box 1), which evaluates how well an ‘ideal’ observer would tell apart two alternative stimulus values by only looking at a response of a single neuron. While this method provides an explicit measure of discrimination per- formance, which is very useful in comparisons of Gauging sensory representations in the brain Giedrius T. Buraˇ cas and Thomas D. Albright The stream of information that enters a sensory system is a product of the ecological niche of an organism and the way in which the information is sampled.The most salient characteristic of this sensory stream is the rich temporal structure that is caused by changes in the environment and self motion of sensors (for example, rapid eye or whisker movements). In recent years, substantial progress has been made in understanding how such rapidly varying stimuli are represented in the responses of sensory neurons of a large variety of sensory systems.The crucial observation that has emerged from these studies is that individual action potentials convey substantial amounts of information, which permits the discrimination of rapidly varying stimuli with high temporal precision. Trends Neurosci. (1999) 22, 303–309 Giedrius T. Buraˇ cas and Thomas D. Albright are at the Howard Hughes Medical Institute, Salk Institute for Biological Studies, La Jolla, CA 92037, USA.