Efficient and adaptive sensory codes

Abstract

Animals exhibit remarkable behavioral flexibility, robustly performing demanding tasks -such as searching for food or avoiding predators-in a variety of different contextual and environmental conditions. However, the demands that detecting and adjusting to changes in the environment place on a sensory system often differ from the demands associated with performing a specific behavioral task, even when both objectives rely on the same sensory modality. This necessitates neural encoding strategies that can dynamically balance these conflicting needs. Here, we develop a theoretical framework that explains how this balance can be achieved, and we use this framework to study tradeoffs in speed, performance, and information transmission that arise as a consequence of efficient coding in dynamic environments. This work generalizes current theories of efficient neural coding to dynamic environments, and thereby provides a unifying perspective on adaptive neural dynamics across different sensory systems, environments, and tasks. this framework to study tradeoffs in speed and performance across a range of different tasks and sensory environments. To our knowledge, this work is the first to provide a normative account of neural coding dynamics underlying adaptive phenomena in the brain. RESULTSA theoretical framework for adaptation. Animals must remain flexible to changes in the environment in order to meet their behavioral demands. As a simple example, consider a system whose goal is to accurately detect the presence of a predator. Different environments might be occupied by different predators, which might in turn be signaled by different distributions of stimulus features ( Fig. 1A). A successful system should have a neural code that can discriminate stimuli within these different distributions. A simple neural code could be constructed by transforming incoming stimuli through a saturating nonlinearity in a manner that approximates the stochastic response function of a sensory neuron ( Fig. 1B). Intuitively, this neuron should use its limited dynamic range to discriminate those stimuli that are most likely to occur in its current environment; this can be achieved by aligning the steep part of its nonlinear response function with the bulk of the incoming stimulus distribution (Fig. 1B, left). This alignment guarantees that the neuron's limited coding capacity is efficiently allocated for the task of accurate stimulus reconstruction, provided that the incoming stimulus distribution does not change in time [26,28,29].In principle, if the environment were to change, the neuron should shift its response function to align with the new distribution in order to maintain accurate performance. In practice, however, the neuron might be insensitive to the stimuli that signal the change, and might then fail to adapt and consequently be unable to distinguish stimuli that signal a predator in the new environment (Fig. 1B, right). Whether and how the neuron adapts to this change, and the impact that this adaptation has on the speed and ...

Dynamics of cortical contrast adaptation predict perception of signals in noise

“…In this study, we directly linked contrast gain control to auditory behavior by combining a theoretical model of efficient coding with simultaneous behavioral psychophysics and recording and manipulation of cortical activity. First, we developed a normative model based on efficient coding 21,23 which predicted that: 1) Detection thresholds of targets should be lower in low contrast than in high contrast; 2) Sensitivity to changes target volume should be greater in low contrast relative to high contrast; and 3) Detection should adapt asymmetrically: increasing slowly after a switch to low contrast, but decreasing rapidly after a switch to high contrast (Figure 1). Then, we used a novel form of Poisson GLM to confirm that gain control dynamics in auditory cortex are indeed asymmetric, as previously observed (Figure 2).…”

“…The efficient coding hypothesis has been formally implemented through normative models of brain function 3,4,[20][21][22][23] . These models assess whether and how neuronal adaptation shapes sensory information and simulate how such adaptation might constrain behavior.…”

“…These models assess whether and how neuronal adaptation shapes sensory information and simulate how such adaptation might constrain behavior. Models based on efficient codes explain psychophysical biases 24 and shape the rate of information transmission when stimulus statistics change dynamically 21,23 . Psychophysical studies suggest that perception is altered by efficient adaptation to stimulus statistics.…”

“…However, because neural and behavioral responses in these studies were not simultaneously measured, a direct relationship between gain control and perceptual performance has yet to be assessed. Additionally, recent theoretical work demonstrated that neuronal dynamics reflect efficient coding 21,23 , but it is unclear whether and how these dynamics shape behavioral performance.…”

Rat sensitivity to multipoint statistics is predicted by efficient coding of natural scenes

“…It is widely believed that the tuning of sensory neurons is determined by the need to efficiently represent the statistical structure of the signals they must encode 1 . This normative principle, known as efficient coding, has been successful in explaining many aspects of neural processing in vision [2][3][4][5][6] , audition 7,8 and olfaction 9 , and in accounting for general mechanisms that are shared across sensory modalities, such as adaptation 10 and gain control 11 . At the perceptual level, strong support for the role played by efficient coding as an organizing principle comes from psychophysical studies showing that human sensitivity to visual textures defined by local multipoint correlations depends on the variability of such correlations across natural scenes 12,13 .…”