Paying attention to natural scenes in area V1.

Natural scene responses in the primary visual cortex are modulated simultaneously by attention and by contextual signals about scene statistics stored across the connectivity of the visual processing hierarchy. We hypothesized that attentional and contextual signals interact in V1 in a manner that primarily benefits the representation of natural stimuli, rich in high-order statistical structure. Recording from two macaques engaged in a spatial attention task, we found that attention enhanced the decodability of stimulus identity from population responses evoked by natural scenes, but not by synthetic stimuli lacking higher-order statistical regularities. Population analysis revealed that neuronal responses converged to a low-dimensional subspace only for natural stimuli. Critically, we determined that the attentional enhancement in stimulus decodability was captured by the natural-scene subspace, indicating an alignment between the attentional and natural stimulus variance. These results suggest that attentional and contextual signals interact in V1 in a manner optimized for natural vision.

Shifts in attention drive context-dependent subspace encoding in anterior cingulate cortex during decision making.

Attention is a cognitive faculty that selects part of a larger set of percepts, driven by cues such as stimulus saliency, internal goals or priors. The enhancement of the attended representation and inhibition of distractors have been proposed as potential neural mechanisms driving this selection process. Yet, how attention operates when the cue has to be internally constructed from conflicting stimuli, decision rules, and reward contingencies, is less understood. Here we recorded from populations of neurons in the anterior cingulate cortex (ACC), an area implicated in ongoing error monitoring and correction during decision conflicts, in a challenging attention-shifting task. In this task, mice had to attend to the rewarded modality when presented identical auditory and visual stimuli in two contexts without direct external cues. In the ACC, the irrelevant stimulus continuously became less decodable than the relevant stimulus as the trial progressed to the decision point. This contrasted strongly with our previous findings in V1 where both relevant and irrelevant stimuli were equally decodable throughout the trial. Using analytical tools and a recurrent neural network (RNN) model, we found that the linearly independent representation of stimulus modalities in ACC was well suited to context-gated suppression of a stimulus modality. We demonstrated that the feedback structure of lateral connections in the RNN consisted of excitatory interactions between cell ensembles representing the same modality and mutual inhibition between cell ensembles representing distinct stimulus modalities. Using this RNN model showing signatures of context-gated suppression, we predicted that the level of contextual modulation of individual neurons should be correlated with their relative responsiveness to the two stimulus modalities used in the task. We verified this prediction in recordings from ACC neurons but not from recordings from V1 neurons. Therefore, ACC effectively operates on low-dimensional neuronal subspaces to combine stimulus related information with internal cues to drive actions under conflict.

Continuous multiplexed population representations of task context in the mouse primary visual cortex.

Effective task execution requires the representation of multiple task-related variables that determine how stimuli lead to correct responses. Even the primary visual cortex (V1) represents other task-related variables such as expectations, choice, and context. However, it is unclear how V1 can flexibly accommodate these variables without interfering with visual representations. We trained mice on a context-switching cross-modal decision task, where performance depends on inferring task context. We found that the context signal that emerged in V1 was behaviorally relevant as it strongly covaried with performance, independent from movement. Importantly, this signal was integrated into V1 representation by multiplexing visual and context signals into orthogonal subspaces. In addition, auditory and choice signals were also multiplexed as these signals were orthogonal to the context representation. Thus, multiplexing allows V1 to integrate visual inputs with other sensory modalities and cognitive variables to avoid interference with the visual representation while ensuring the maintenance of task-relevant variables.

Sampling motion trajectories during hippocampal theta sequences.

Efficient planning in complex environments requires that uncertainty associated with current inferences and possible consequences of forthcoming actions is represented. Representation of uncertainty has been established in sensory systems during simple perceptual decision making tasks but it remains unclear if complex cognitive computations such as planning and navigation are also supported by probabilistic neural representations. Here, we capitalized on gradually changing uncertainty along planned motion trajectories during hippocampal theta sequences to capture signatures of uncertainty representation in population responses. In contrast with prominent theories, we found no evidence of encoding parameters of probability distributions in the momentary population activity recorded in an open-field navigation task in rats. Instead, uncertainty was encoded sequentially by sampling motion trajectories randomly and efficiently in subsequent theta cycles from the distribution of potential trajectories. Our analysis is the first to demonstrate that the hippocampus is well equipped to contribute to optimal planning by representing uncertainty.

Tracking the contribution of inductive bias to individualised internal models.

Internal models capture the regularities of the environment and are central to understanding how humans adapt to environmental statistics. In general, the correct internal model is unknown to observers, instead they rely on an approximate model that is continually adapted throughout learning. However, experimenters assume an ideal observer model, which captures stimulus structure but ignores the diverging hypotheses that humans form during learning. We combine non-parametric Bayesian methods and probabilistic programming to infer rich and dynamic individualised internal models from response times. We demonstrate that the approach is capable of characterizing the discrepancy between the internal model maintained by individuals and the ideal observer model and to track the evolution of the contribution of the ideal observer model to the internal model throughout training. In particular, in an implicit visuomotor sequence learning task the identified discrepancy revealed an inductive bias that was consistent across individuals but varied in strength and persistence.

Optimal forgetting: Semantic compression of episodic memories.

It has extensively been documented that human memory exhibits a wide range of systematic distortions, which have been associated with resource constraints. Resource constraints on memory can be formalised in the normative framework of lossy compression, however traditional lossy compression algorithms result in qualitatively different distortions to those found in experiments with humans. We argue that the form of distortions is characteristic of relying on a generative model adapted to the environment for compression. We show that this semantic compression framework can provide a unifying explanation of a wide variety of memory phenomena. We harness recent advances in learning deep generative models, that yield powerful tools to approximate generative models of complex data. We use three datasets, chess games, natural text, and hand-drawn sketches, to demonstrate the effects of semantic compression on memory performance. Our model accounts for memory distortions related to domain expertise, gist-based distortions, contextual effects, and delayed recall.

Representational untangling by the firing rate nonlinearity in V1 simple cells.

An important computational goal of the visual system is 'representational untangling' (RU): representing increasingly complex features of visual scenes in an easily decodable format. RU is typically assumed to be achieved in high-level visual cortices via several stages of cortical processing. Here we show, using a canonical population coding model, that RU of low-level orientation information is already performed at the first cortical stage of visual processing, but not before that, by a fundamental cellular-level property: the thresholded firing rate nonlinearity of simple cells in the primary visual cortex (V1). We identified specific, experimentally measurable parameters that determined the optimal firing threshold for RU and found that the thresholds of V1 simple cells extracted from in vivo recordings in awake behaving mice were near optimal. These results suggest that information re-formatting, rather than maximisation, may already be a relevant computational goal for the early visual system.

Stimulus complexity shapes response correlations in primary visual cortex.

Spike count correlations (SCCs) are ubiquitous in sensory cortices, are characterized by rich structure, and arise from structured internal dynamics. However, most theories of visual perception treat contributions of neurons to the representation of stimuli independently and focus on mean responses. Here, we argue that, in a functional model of visual perception, featuring probabilistic inference over a hierarchy of features, inferences about high-level features modulate inferences about low-level features ultimately introducing structured internal dynamics and patterns in SCCs. Specifically, high-level inferences for complex stimuli establish the local context in which neurons in the primary visual cortex (V1) interpret stimuli. Since the local context differentially affects multiple neurons, this conjecture predicts specific modulations in the fine structure of SCCs as stimulus identity and, more importantly, stimulus complexity varies. We designed experiments with natural and synthetic stimuli to measure the fine structure of SCCs in V1 of awake behaving macaques and assessed their dependence on stimulus identity and stimulus statistics. We show that the fine structure of SCCs is specific to the identity of natural stimuli and changes in SCCs are independent of changes in response mean. Critically, we demonstrate that stimulus specificity of SCCs in V1 can be directly manipulated by altering the amount of high-order structure in synthetic stimuli. Finally, we show that simple phenomenological models of V1 activity cannot account for the observed SCC patterns and conclude that the stimulus dependence of SCCs is a natural consequence of structured internal dynamics in a hierarchical probabilistic model of natural images.

Measuring and filtering reactive inhibition is essential for assessing serial decision making and learning.

Learning complex structures from stimuli requires extended exposure and often repeated observation of the same stimuli. Learning induces stimulus-dependent changes in specific performance measures. The same performance measures, however, can also be affected by processes that arise because of extended training (e.g., fatigue) but are otherwise independent from learning. Thus, a thorough assessment of the properties of learning can only be achieved by identifying and accounting for the effects of such processes. Reactive inhibition is a process that modulates behavioral performance measures on a wide range of time scales and often has opposite effects than learning. Here we develop a tool to disentangle the effects of reactive inhibition from learning in the context of an implicit learning task, the alternating serial reaction time (ASRT) task. Our method highlights that the magnitude of the effect of reactive inhibition on measured performance is larger than that of the acquisition of statistical structure from stimuli. We show that the effect of reactive inhibition can be identified not only in population measures but also at the level of performance of individuals, revealing varying degrees of contribution of reactive inhibition. Finally, we demonstrate that a higher proportion of behavioral variance can be explained by learning once the effects of reactive inhibition are eliminated. These results demonstrate that reactive inhibition has a fundamental effect on the behavioral performance that can be identified in individual participants and can be separated from other cognitive processes like learning. (PsycINFO Database Record

Population activity statistics dissect subthreshold and spiking variability in V1.

Response variability, as measured by fluctuating responses upon repeated performance of trials, is a major component of neural responses, and its characterization is key to interpret high dimensional population recordings. Response variability and covariability display predictable changes upon changes in stimulus and cognitive or behavioral state, providing an opportunity to test the predictive power of models of neural variability. Still, there is little agreement on which model to use as a building block for population-level analyses, and models of variability are often treated as a subject of choice. We investigate two competing models, the doubly stochastic Poisson (DSP) model assuming stochasticity at spike generation, and the rectified Gaussian (RG) model tracing variability back to membrane potential variance, to analyze stimulus-dependent modulation of both single-neuron and pairwise response statistics. Using a pair of model neurons, we demonstrate that the two models predict similar single-cell statistics. However, DSP and RG models have contradicting predictions on the joint statistics of spiking responses. To test the models against data, we build a population model to simulate stimulus change-related modulations in pairwise response statistics. We use single-unit data from the primary visual cortex (V1) of monkeys to show that while model predictions for variance are qualitatively similar to experimental data, only the RG model's predictions are compatible with joint statistics. These results suggest that models using Poisson-like variability might fail to capture important properties of response statistics. We argue that membrane potential-level modeling of stochasticity provides an efficient strategy to model correlations.NEW & NOTEWORTHY Neural variability and covariability are puzzling aspects of cortical computations. For efficient decoding and prediction, models of information encoding in neural populations hinge on an appropriate model of variability. Our work shows that stimulus-dependent changes in pairwise but not in single-cell statistics can differentiate between two widely used models of neuronal variability. Contrasting model predictions with neuronal data provides hints on the noise sources in spiking and provides constraints on statistical models of population activity.

Neural Variability and Sampling-Based Probabilistic Representations in the Visual Cortex.

Neural responses in the visual cortex are variable, and there is now an abundance of data characterizing how the magnitude and structure of this variability depends on the stimulus. Current theories of cortical computation fail to account for these data; they either ignore variability altogether or only model its unstructured Poisson-like aspects. We develop a theory in which the cortex performs probabilistic inference such that population activity patterns represent statistical samples from the inferred probability distribution. Our main prediction is that perceptual uncertainty is directly encoded by the variability, rather than the average, of cortical responses. Through direct comparisons to previously published data as well as original data analyses, we show that a sampling-based probabilistic representation accounts for the structure of noise, signal, and spontaneous response variability and correlations in the primary visual cortex. These results suggest a novel role for neural variability in cortical dynamics and computations.

Cognitive tomography reveals complex, task-independent mental representations.

Humans develop rich mental representations that guide their behavior in a variety of everyday tasks. However, it is unknown whether these representations, often formalized as priors in Bayesian inference, are specific for each task or subserve multiple tasks. Current approaches cannot distinguish between these two possibilities because they cannot extract comparable representations across different tasks. Here, we develop a novel method, termed cognitive tomography, that can extract complex, multidimensional priors across tasks. We apply this method to human judgments in two qualitatively different tasks, "familiarity" and "odd one out," involving an ecologically relevant set of stimuli, human faces. We show that priors over faces are structurally complex and vary dramatically across subjects, but are invariant across the tasks within each subject. The priors we extract from each task allow us to predict with high precision the behavior of subjects for novel stimuli both in the same task as well as in the other task. Our results provide the first evidence for a single high-dimensional structured representation of a naturalistic stimulus set that guides behavior in multiple tasks. Moreover, the representations estimated by cognitive tomography can provide independent, behavior-based regressors for elucidating the neural correlates of complex naturalistic priors.

Representations of uncertainty in sensorimotor control.

Uncertainty is ubiquitous in our sensorimotor interactions, arising from factors such as sensory and motor noise and ambiguity about the environment. Setting it apart from previous theories, a quintessential property of the Bayesian framework for making inference about the state of world so as to select actions, is the requirement to represent the uncertainty associated with inferences in the form of probability distributions. In the context of sensorimotor control and learning, the Bayesian framework suggests that to respond optimally to environmental stimuli the central nervous system needs to construct estimates of the sensorimotor transformations, in the form of internal models, as well as represent the structure of the uncertainty in the inputs, outputs and in the transformations themselves. Here we review Bayesian inference and learning models that have been successful in demonstrating the sensitivity of the sensorimotor system to different forms of uncertainty as well as recent studies aimed at characterizing the representation of the uncertainty at different computational levels.

Spontaneous cortical activity reveals hallmarks of an optimal internal model of the environment.

The brain maintains internal models of its environment to interpret sensory inputs and to prepare actions. Although behavioral studies have demonstrated that these internal models are optimally adapted to the statistics of the environment, the neural underpinning of this adaptation is unknown. Using a Bayesian model of sensory cortical processing, we related stimulus-evoked and spontaneous neural activities to inferences and prior expectations in an internal model and predicted that they should match if the model is statistically optimal. To test this prediction, we analyzed visual cortical activity of awake ferrets during development. Similarity between spontaneous and evoked activities increased with age and was specific to responses evoked by natural scenes. This demonstrates the progressive adaptation of internal models to the statistics of natural stimuli at the neural level.

Statistically optimal perception and learning: from behavior to neural representations.

Human perception has recently been characterized as statistical inference based on noisy and ambiguous sensory inputs. Moreover, suitable neural representations of uncertainty have been identified that could underlie such probabilistic computations. In this review, we argue that learning an internal model of the sensory environment is another key aspect of the same statistical inference procedure and thus perception and learning need to be treated jointly. We review evidence for statistically optimal learning in humans and animals, and re-evaluate possible neural representations of uncertainty based on their potential to support statistically optimal learning. We propose that spontaneous activity can have a functional role in such representations leading to a new, sampling-based, framework of how the cortex represents information and uncertainty.

Bayesian learning of visual chunks by human observers.

Efficient and versatile processing of any hierarchically structured information requires a learning mechanism that combines lower-level features into higher-level chunks. We investigated this chunking mechanism in humans with a visual pattern-learning paradigm. We developed an ideal learner based on Bayesian model comparison that extracts and stores only those chunks of information that are minimally sufficient to encode a set of visual scenes. Our ideal Bayesian chunk learner not only reproduced the results of a large set of previous empirical findings in the domain of human pattern learning but also made a key prediction that we confirmed experimentally. In accordance with Bayesian learning but contrary to associative learning, human performance was well above chance when pair-wise statistics in the exemplars contained no relevant information. Thus, humans extract chunks from complex visual patterns by generating accurate yet economical representations and not by encoding the full correlational structure of the input.

Pharmacological and computational analysis of alpha-subunit preferential GABA(A) positive allosteric modulators on the rat septo-hippocampal activity.

Clinically most active anxiolytic drugs are positive allosteric modulators (PAMs) of GABA(A) receptors, represented by benzodiazepine compounds. Due to their non-selective profile, however, they potently modulate several sup-type specific GABA(A) receptors, contributing to their broad-range side effects. Based on observations in genetically altered mice, however, it has been proposed that anxiolytic action of benzodiazepines is predominantly mediated by GABA(A) alpha2/3 subunit-containing receptors. In the present study we analyzed the actions of the preferential GABA(A) alpha1 and alpha2/3 PAMs, zolpidem and L-838417, respectively on hippocampal EEG and medial septum neuronal activity in anesthetized rats. In parallel, a computational model was constructed to model pharmacological actions of these compounds on the septo-hippocampal circuitry. The present results demonstrated that zolpidem inhibited theta oscillation both in the hippocampus and septum, and profoundly inhibited firing activity of septal neurons. L-838417 also inhibited hippocampal and septal theta oscillation, however, it did not significantly alter firing rate activity of septal neurons. Our computational model showed that cessation of periodic firing of hippocampo-septal neurons, representing absence of hippocampal theta activity, disrupted oscillation of septal units, without altering their overall firing activity, similar to changes observed in our in vivo experiments following administration of L-838417. Understanding the correlation between changes in septo-hippocampal activity and actions of selective modulators of GABA(A) subtype receptor modulators would further advance design of anxiolytic drugs.

Intrinsic and synaptic mechanisms determining the timing of neuron population activity during hippocampal theta oscillation.

Hippocampal theta (3-8 Hz) is a major electrophysiological activity in rodents, which can be found in primates and humans as well. During theta activity, pyramidal cells and different classes of interneurons were shown to discharge at different phases of the extracellular theta. A recent in vitro study has shown that theta-frequency oscillation can be elicited in a hippocampal CA1 slice by the activation of metabotropic glutamate receptors with similar pharmacological and physiological profile that was found in vivo. We constructed a conductance based three-population network model of the hippocampal CA1 region to study the specific roles of neuron types in the generation of the in vitro theta oscillation and the emergent network properties. Interactions between pairs of neuron populations were studied systematically to assess synchronization and delay properties. We showed that the circuitry consisting of pyramidal cells and two types of hippocampal interneurons [basket and oriens lacunosum-moleculare (O-LM) neurons] was able to generate coherent theta-frequency population oscillation. Furthermore, we found that hyperpolarization-activated nonspecific cation current in pyramidal cells, but not in O-LM neurons, plays an important role in the timing of spike generation, and thus synchronization of pyramidal cells. The model was shown to exhibit the same phase differences between neuron population activities found in vivo, supporting the idea that these patterns of activity are determined internal to the hippocampus.

Theta oscillation-coupled dendritic spiking integrates inputs on a long time scale.

Persistent neural activity lasting for seconds after transient stimulation has been observed in several brain areas. This activity has been taken to be indicative of the integration of inputs on long time scales. Passive membrane properties render neural time constants to be on the order of milliseconds. Intense synaptic bombardment, characteristic of in vivo states, was previously shown to further reduce the time scale of effective integration. We explored how long-term integration in single cells could be supported by dendritic spikes coupled with the theta oscillation, a prominent brain rhythm often observed during working memory tasks. We used a two-compartmental conductance-based model of a hippocampal pyramidal cell to study the interplay of intrinsic dynamics with periodic inputs in the theta frequency band. We show that periodic dendritic spiking integrates inputs by shifting the phase relative to an external oscillation, since spiking frequency is quasi-linearly modulated by current injection. The time-constant of this integration process is practically infinite for input intensities above a threshold (the integration threshold) and can be still several hundred milliseconds long below the integration threshold. The somatic compartment received theta frequency stimulation in antiphase with the dendritic oscillation. Consequently, dendritic spikes could only elicit somatic action potentials when they were sufficiently phase-shifted and thus coincided with somatic depolarization. Somatic depolarization modulated the frequency but not the phase of firing, endowing the cell with the capability to code for two different variables at the same time. Inputs to the dendrite shifted the phase of dendritic spiking, while somatic input was modulating its firing rate. This mechanism resulted in firing patterns that closely matched experimental data from hippocampal place cells of freely behaving rats. We discuss the plausibility of our proposed mechanism and its potential to account for the firing pattern of cells outside the hippocampus during working memory tasks.