Perceptual decisions are based on sensory information encoded by populations of neurons over short periods. To identify the most important aspects of the population code for guiding behavior, I varied visual attention, which improves perception of an attended location or feature. The amount of information encoded in a population of neurons is limited by both the variability of individual cortical neurons and by correlated variability that is shared across the population (noise correlations). I found that attentional changes in shared variability are likely the major contributor to the resulting behavioral improvement. These results suggest that studies of single neurons that ignore noise correlations miss perhaps the most crucial aspect of the way that responses of populations of neurons guide behavior, so it is necessary to monitor the activity of populations of neurons rather than record one neuron at a time. The goal of this proposal is to use computational methods and multielectrode recordings in awake primates to understand the mechanism by which inter-neuronal correlations arise and the extent to which they can flexibly adapt to task demands. In my previous experiments I found that attentional modulation of firing rates and noise correlations is linked: the neurons that show the biggest firing rate changes (typically increases in multiplicative gain) show the biggest decreases in correlation. I will use the mentored phase of this award to investigate the origin and flexibility of noise correlations by creating a computational model to test whether a mechanism thought to underlie gain changes in single neurons in many sensory, motor and cognitive processes (sensory normalization) could also cause modulation of noise correlations (Aim 1). The result that modulation of correlations accompanies all gain changes would imply that correlation changes are a major factor in most cortical computations. My goal in the independent phase will be to experimentally investigate how flexible and adaptive correlation changes can be and the impact of correlations on communication between cortical areas. Noise correlations can either severely limit or improve the information available in a neuronal population, depending on the way that neuronal responses are combined. In the task I used in my previous experiments, noise correlations limited population sensitivity, and attention adaptively decreased correlations (as predicted by the normalization hypothesis).
In Aim 2, I will use a task in which noise correlations improve, rather than limit, performance to determine whether attention can adaptively increase correlations.
In Aim 3, I will investigate the origin of noise correlations by measuring correlations between different cortical areas and varying attention to determine whether correlations depend on the strength of shared functional inputs. Collectively, these studies will have implications for the impact of correlations on population coding and the way sensory information is transmitted from area to area and used to guide behavior.
Many neurological diseases including depression, schizophrenia, and attention deficit hyperactivity disorder (ADHD) are thought to involve networks of cortical neurons, so understanding the way that information is encoded and read out by neuronal populations will be critical for diagnosing and developing drug therapies to treat these diseases. The projects in this proposal will elucidate general mechanisms for population coding, and will in particular enhance our understanding of the neuronal mechanisms underlying attention. Deficits of attention such as ADHD are thought to affect as many as 5% of the children in the United States, and better understanding of the basic neuronal mechanisms related to attention is needed for guiding assessment, diagnosis, and treatment of attentional deficits.
Showing the most recent 10 out of 12 publications
