22When making a categorical decision about a noisy stimulus, it is common to fluctuate between 44 levels of commitment to a choice before reporting a decision. In some instances the fluctuations 45 are sufficiently strong to lead to a "change of mind" (CoM) while deliberating [1][2][3][4][5][6] or even while the 46 reporting action is being executed 7 . Because these within-trial fluctuations are different from trial 47 to trial and not necessarily tied to an external event or stimulus feature, they can only be captured 48 using a moment-to-moment neural readout of the decision state on single trials. 49To obtain this readout, we decoded a decision variable (DV) from neural population activity in 50 PMd and M1 in real time to continuously estimate the decision state while two monkeys performed 51 a motion discrimination task 8,9 (Fig. 1a, see Methods). The DV was estimated by applying a linear 52 decoder, trained on data from a previous experimental session, to spiking data (from 96 to 192 53 electrodes) from the preceding 50 ms, updated every 10 ms throughout each trial ( Fig. 1b, see 54 Methods). The sign of the DV indicated which choice was predicted by the decoder, which allowed 55 us to calculate the decoder's prediction accuracy. The DV magnitude reflected the confidence of 56 the model's prediction in units of log-odds for one vs. the other decision (see Methods). Note that 57 the decision variable as defined here encompasses all choice predictive signals that can be decoded 58 from neural activity 10 , including but not limited to moment-to-moment value of accumulated 59 evidence as posited in classical sequential sampling models. 60We have previously demonstrated with offline analysis that this decision variable (DV) can predict 61 choices on single trials up to seconds before initiation of the operant response, and that the 62 accuracy of these predictions increases on average throughout the course of the trial 10 . 63Here, we employed closed-loop, neurally-contingent control over stimulus timing to directly probe 64 the relationship of within-trial DV fluctuations to behaviorally meaningful decision states. For the 65 4 first time, we quantified the behavioral effects of previously covert DV variations (i) as a function 66 of time and for different virtual DV boundaries imposed during the trial, (ii) when large, CoM-like 67 fluctuations were detected during deliberation on noisy visual evidence, and (iii) when 68 subthreshold stimulus pulses were added during the trial. 69Having a nearly instantaneous real-time estimate of the decision state read-out enabled us to 70 terminate the visual stimulus based on the current value (or history) of the DV and validate the 71 behavioral relevance of DV fluctuations using the monkey's behavioral reports following stimulus 72 termination. 73Decisions on perceived stimulus motion can be reliably decoded in real time based on 50 ms 74 of PMd/M1 neural activity 75 76 Two monkeys performed a variable duration variant of the classical random dot motion 77 discriminatio...

In dynamic environments, subjects often integrate multiple samples of a signal and combine them to reach a categorical judgment. The process of deliberation on the evidence can be described by a time-varying decision variable (DV), decoded from neural activity, that predicts a subject's decision at the end of a trial. However, within trials, large moment-to-moment fluctuations of the DV are observed. The behavioral significance of these fluctuations and their role in the decision process remain unclear. Here we show that within-trial DV fluctuations decoded in real time from motor cortex are tightly linked to choice behavior, and that robust changes in DV sign have the statistical regularities expected from behavioral studies of changes-of-mind. Furthermore, we find single-trial evidence for absorbing decision bounds. As the DV builds up, heavily favoring one or the other choice, moment-to-moment variability in the DV is reduced, and both neural DV and behavioral decisions become resistant to additional pulses of sensory evidence as predicted by diffusion-to-bound and attractor models of the decision process.

Sensory environments often contain an overwhelming amount of information, with both relevant and irrelevant information competing for neural resources. Feature attention mediates this competition by selecting the sensory features needed to form a coherent percept. How attention affects the activity of populations of neurons to support this process is poorly understood because population coding is typically studied through simulations in which one sensory feature is encoded without competition. Therefore, to study the effects of feature attention on population-based neural coding, investigations must be extended to include stimuli with both relevant and irrelevant features. We measured noise correlations () within small neural populations in primary auditory cortex while rhesus macaques performed a novel feature-selective attention task. We found that the effect of feature-selective attention on depended not only on the population tuning to the attended feature, but also on the tuning to the distractor feature. To attempt to explain how these observed effects might support enhanced perceptual performance, we propose an extension of a simple and influential model in which shifts in can simultaneously enhance the representation of the attended feature while suppressing the distractor. These findings present a novel mechanism by which attention modulates neural populations to support sensory processing in cluttered environments. Although feature-selective attention constitutes one of the building blocks of listening in natural environments, its neural bases remain obscure. To address this, we developed a novel auditory feature-selective attention task and measured noise correlations () in rhesus macaque A1 during task performance. Unlike previous studies showing that the effect of attention on depends on population tuning to the attended feature, we show that the effect of attention depends on the tuning to the distractor feature as well. We suggest that these effects represent an efficient process by which sensory cortex simultaneously enhances relevant information and suppresses irrelevant information.

Textbook descriptions of primary sensory cortex (PSC) revolve around single neurons' representation of low-dimensional sensory features, such as visual object orientation in V1, location of somatic touch in S1, and sound frequency in A1. Typically, studies of PSC measure neurons' responses along few (1 or 2) stimulus and/or behavioral dimensions. However, real-world stimuli usually vary along many feature dimensions and behavioral demands change constantly. In order to illuminate how A1 supports flexible perception in rich acoustic environments, we recorded from A1 neurons while rhesus macaques performed a feature-selective attention task. We presented sounds that varied along spectral and temporal feature dimensions (carrier bandwidth and temporal envelope, respectively). Within a block, subjects attended to one feature of the sound in a selective change detection task. We find that single neurons tend to be highdimensional, in that they exhibit substantial mixed selectivity for both sound features, as well as task context. Contrary to common findings in many previous experiments, attention does not enhance the single-neuron representation of attended features in our data. However, a population-level analysis reveals that ensembles of neurons exhibit enhanced encoding of attended sound features, and this population code tracks subjects' performance. Importantly, surrogate neural populations with intact singleneuron tuning but shuffled higher-order correlations among neurons failed to yield attention-related effects observed in the intact data. These results suggest that an emergent population code not measurable at the single-neuron level might constitute the functional unit of sensory representation in PSC. SIGNIFICANCE STATEMENTThe ability to adapt to a dynamic sensory environment promotes a range of important natural behaviors. We recorded from single neurons in monkey primary auditory cortex while subjects attended to either the spectral or temporal features of complex sounds.Surprisingly, we find no average increase in responsiveness to, or encoding of, the attended feature across single neurons. However, when we pool the activity of the sampled neurons via targeted dimensionality reduction, we find enhanced populationlevel representation of the attended feature and suppression of the distractor feature.This dissociation of the effects of attention at the level of single neurons vs. the population highlights the synergistic nature of cortical sound encoding and enriches our understanding of sensory cortical function.