Attention and Visual Perception
Attention is critical for optimal behavior. We study how neurons change the way they represented the visual world when attention shifts from one object to another. Typically, signals from neurons that encode visual information are stronger and less noisy when a subject pays attention to the objects that those neurons represent. Our current projects are directed at two important questions about the neuronal correlates of attention.
First, what is attention? Studies of attention have always used operational definitions. Attention takes different forms, and we have been working to break attention down into well defined components that map onto contributions from different brain structures. We have found that attention-related changes in the signals of neurons in area V4 in visual cortex are specifically related to behavioral improvements in the ability to discriminate stimuli, while changes in prefrontal cortex are related to changes in preparation to respond to target stimuli (Luo & Maunsell, 2015, 2018, 2019). We have also found that neurons in area V4 modulate their signals when subjects pay more or less attention to a given object (Ghosh & Maunsell, 2021).
Our current experiments are examining how sub-cortical structures like the superior colliculus and locus coeruleus contribute to visual attention, and the role of specific neuromodulatory systems to specific dimensions of attention.
Normalization in Sensory Representations
Sensory neurons respond differently when two or more stimuli appear at once. Typically, a neuron's response to multiple stimuli is not simply the sum of its responses to those stimuli when they appear individually. Instead, responses are normalized to the context or how many stimuli are present. This produces a response that is closer to the average of the responses to the individual stimuli.
Normalization is ubiquitous in the nervous system. While normalization is undoubtedly important for best using the dynamic range of neuronal responses, evidence suggests it serves other important functions. In particular, it appears that much of the attenton-related boost in sensory signals depends on normalization circuits to amplify signals (Ni & Maunsell, 2019). Normalization circuits can also reduce the noise in population signals (Verhoef & Maunsell 2017). Additionally, normalization circuits might contribute to oscillations in the gamma frequency range in the local field potential (Ray et al. 2013). We are using multielectrode recordings in mice and monkeys to characterize the specific cell-types that contribute to normalization, the circuits that they form, and the range of functional effects that they have on information processing in the brain.
Readout of Signals from Cerebral Cortex
The cerebral cortex provides the brain's best and most complete representation of its environment and state: sensory stimuli, current goals, task rules, cognitive set, motor state and more are all represented in cerebral cortex. At any moment, some of this information will be absolutely critical while much will be completely irrelevant. Which representations are high-priority can change from moment to moment. Working out how the brain selects and accesses specific cortical representations is one of the great challenges in systems-level neuroscience. We currently have no comprehensive answers to questions such as how widely or narrowly cortical signals can be integrated in space and time, whether fine temporal patterns of spike in cortex convey useful information, or how frequently and rapidly readout can shift from one set of neurons to another.
We have been approaching this issue using optogenetic perturbation of cortical neurons in mice that have been trained to perform visual detection and discrimination tasks (Cone et al, 2019, 2020). Having animals give behavioral reports provides unambiguous evidence that the signals can guide behavior (i.e., they can be read out). Because the relevant signals are those we introduce, we know their location(s) and temporal characteristics precisely. Using optogenetics, we can limit perturbations to genetically identified sub-types of neurons. We have shown that mice are better at detecting increases in cortical spike rates than decreases in cortical spike rates (Cone 2020). We have recently completely human psychophysical experiments that suggests this asymmetry in decoding exists across all mammals (Wei et al. 2023).
