Multitasking via baseline control in recurrent neural networks.

Changes in behavioral state, such as arousal and movements, strongly affect neural activity in sensory areas, and can be modeled as long-range projections regulating the mean and variance of baseline input currents. What are the computational benefits of these baseline modulations? We investigate this question within a brain-inspired framework for reservoir computing, where we vary the quenched baseline inputs to a recurrent neural network with random couplings. We found that baseline modulations control the dynamical phase of the reservoir network, unlocking a vast repertoire of network phases. We uncovered a number of bistable phases exhibiting the simultaneous coexistence of fixed points and chaos, of two fixed points, and of weak and strong chaos. We identified several phenomena, including noise-driven enhancement of chaos and ergodicity breaking; neural hysteresis, whereby transitions across a phase boundary retain the memory of the preceding phase. In each bistable phase, the reservoir performs a different binary decision-making task. Fast switching between different tasks can be controlled by adjusting the baseline input mean and variance. Moreover, we found that the reservoir network achieves optimal memory performance at any first-order phase boundary. In summary, baseline control enables multitasking without any optimization of the network couplings, opening directions for brain-inspired artificial intelligence and providing an interpretation for the ubiquitously observed behavioral modulations of cortical activity.

State-Dependent Regulation of Cortical Processing Speed via Gain Modulation.

To thrive in dynamic environments, animals must be capable of rapidly and flexibly adapting behavioral responses to a changing context and internal state. Examples of behavioral flexibility include faster stimulus responses when attentive and slower responses when distracted. Contextual or state-dependent modulations may occur early in the cortical hierarchy and may be implemented via top-down projections from corticocortical or neuromodulatory pathways. However, the computational mechanisms mediating the effects of such projections are not known. Here, we introduce a theoretical framework to classify the effects of cell type-specific top-down perturbations on the information processing speed of cortical circuits. Our theory demonstrates that perturbation effects on stimulus processing can be predicted by intrinsic gain modulation, which controls the timescale of the circuit dynamics. Our theory leads to counterintuitive effects, such as improved performance with increased input variance. We tested the model predictions using large-scale electrophysiological recordings from the visual hierarchy in freely running mice, where we found that a decrease in single-cell intrinsic gain during locomotion led to an acceleration of visual processing. Our results establish a novel theory of cell type-specific perturbations, applicable to top-down modulation as well as optogenetic and pharmacological manipulations. Our theory links connectivity, dynamics, and information processing via gain modulation.SIGNIFICANCE STATEMENT To thrive in dynamic environments, animals adapt their behavior to changing circumstances and different internal states. Examples of behavioral flexibility include faster responses to sensory stimuli when attentive and slower responses when distracted. Previous work suggested that contextual modulations may be implemented via top-down inputs to sensory cortex coming from higher brain areas or neuromodulatory pathways. Here, we introduce a theory explaining how the speed at which sensory cortex processes incoming information is adjusted by changes in these top-down projections, which control the timescale of neural activity. We tested our model predictions in freely running mice, revealing that locomotion accelerates visual processing. Our theory is applicable to internal modulation as well as optogenetic and pharmacological manipulations and links circuit connectivity, dynamics, and information processing.

Dynamics of multistable states during ongoing and evoked cortical activity.

Single-trial analyses of ensemble activity in alert animals demonstrate that cortical circuits dynamics evolve through temporal sequences of metastable states. Metastability has been studied for its potential role in sensory coding, memory, and decision-making. Yet, very little is known about the network mechanisms responsible for its genesis. It is often assumed that the onset of state sequences is triggered by an external stimulus. Here we show that state sequences can be observed also in the absence of overt sensory stimulation. Analysis of multielectrode recordings from the gustatory cortex of alert rats revealed ongoing sequences of states, where single neurons spontaneously attain several firing rates across different states. This single-neuron multistability represents a challenge to existing spiking network models, where typically each neuron is at most bistable. We present a recurrent spiking network model that accounts for both the spontaneous generation of state sequences and the multistability in single-neuron firing rates. Each state results from the activation of neural clusters with potentiated intracluster connections, with the firing rate in each cluster depending on the number of active clusters. Simulations show that the model's ensemble activity hops among the different states, reproducing the ongoing dynamics observed in the data. When probed with external stimuli, the model predicts the quenching of single-neuron multistability into bistability and the reduction of trial-by-trial variability. Both predictions were confirmed in the data. Together, these results provide a theoretical framework that captures both ongoing and evoked network dynamics in a single mechanistic model.

Processing of hedonic and chemosensory features of taste in medial prefrontal and insular networks.

Most of the research on cortical processing of taste has focused on either the primary gustatory cortex (GC) or the orbitofrontal cortex (OFC). However, these are not the only areas involved in taste processing. Gustatory information can also reach another frontal region, the medial prefrontal cortex (mPFC), via direct projections from GC. mPFC has been studied extensively in relation to its role in controlling goal-directed action and reward-guided behaviors, yet very little is known about its involvement in taste coding. The experiments presented here address this important point and test whether neurons in mPFC can significantly process the physiochemical and hedonic dimensions of taste. Spiking responses to intraorally delivered tastants were recorded from rats implanted with bundles of electrodes in mPFC and GC. Analysis of single-neuron and ensemble activity revealed similarities and differences between the two areas. Neurons in mPFC can encode the chemosensory identity of gustatory stimuli. However, responses in mPFC are sparser, more narrowly tuned, and have a later onset than in GC. Although taste quality is more robustly represented in GC, taste palatability is coded equally well in the two areas. Additional analysis of responses in neurons processing the hedonic value of taste revealed differences between the two areas in temporal dynamics and sensitivities to palatability. These results add mPFC to the network of areas involved in processing gustatory stimuli and demonstrate significant differences in taste-coding between GC and mPFC.

Decision-making dynamics are predicted by arousal and uninstructed movements.

During sensory-guided behavior, an animal's decision-making dynamics unfold through sequences of distinct performance states, even while stimulus-reward contingencies remain static. Little is known about the factors that underlie these changes in task performance. We hypothesize that these decision-making dynamics can be predicted by externally observable measures, such as uninstructed movements and changes in arousal. Here, using computational modeling of visual and auditory task performance data from mice, we uncovered lawful relationships between transitions in strategic task performance states and an animal's arousal and uninstructed movements. Using hidden Markov models applied to behavioral choices during sensory discrimination tasks, we find that animals fluctuate between minutes-long optimal, sub-optimal, and disengaged performance states. Optimal state epochs are predicted by intermediate levels, and reduced variability, of pupil diameter and movement. Our results demonstrate that externally observable uninstructed behaviors can predict optimal performance states and suggest that mice regulate their arousal during optimal performance.

Modulation of metastable ensemble dynamics explains optimal coding at moderate arousal in auditory cortex.

Performance during perceptual decision-making exhibits an inverted-U relationship with arousal, but the underlying network mechanisms remain unclear. Here, we recorded from auditory cortex (A1) of behaving mice during passive tone presentation, while tracking arousal via pupillometry. We found that tone discriminability in A1 ensembles was optimal at intermediate arousal, revealing a population-level neural correlate of the inverted-U relationship. We explained this arousal-dependent coding using a spiking network model with a clustered architecture. Specifically, we show that optimal stimulus discriminability is achieved near a transition between a multi-attractor phase with metastable cluster dynamics (low arousal) and a single-attractor phase (high arousal). Additional signatures of this transition include arousal-induced reductions of overall neural variability and the extent of stimulus-induced variability quenching, which we observed in the empirical data. Altogether, this study elucidates computational principles underlying interactions between pupil-linked arousal, sensory processing, and neural variability, and suggests a role for phase transitions in explaining nonlinear modulations of cortical computations.

Probabilistic modeling reveals coordinated social interaction states and their multisensory bases.

Social behavior across animal species ranges from simple pairwise interactions to thousands of individuals coordinating goal-directed movements. Regardless of the scale, these interactions are governed by the interplay between multimodal sensory information and the internal state of each animal. Here, we investigate how animals use multiple sensory modalities to guide social behavior in the highly social zebrafish (Danio rerio) and uncover the complex features of pairwise interactions early in development. To identify distinct behaviors and understand how they vary over time, we developed a new hidden Markov model with constrained linear-model emissions to automatically classify states of coordinated interaction, using the movements of one animal to predict those of another. We discovered that social behaviors alternate between two interaction states within a single experimental session, distinguished by unique movements and timescales. Long-range interactions, akin to shoaling, rely on vision, while mechanosensation underlies rapid synchronized movements and parallel swimming, precursors of schooling. Altogether, we observe spontaneous interactions in pairs of fish, develop novel hidden Markov modeling to reveal two fundamental interaction modes, and identify the sensory systems involved in each. Our modeling approach to pairwise social interactions has broad applicability to a wide variety of naturalistic behaviors and species and solves the challenge of detecting transient couplings between quasi-periodic time series.

Expressivity of Neural Networks with Random Weights and Learned Biases.

Landmark universal function approximation results for neural networks with trained weights and biases provided impetus for the ubiquitous use of neural networks as learning models in Artificial Intelligence (AI) and neuroscience. Recent work has pushed the bounds of universal approximation by showing that arbitrary functions can similarly be learned by tuning smaller subsets of parameters, for example the output weights, within randomly initialized networks. Motivated by the fact that biases can be interpreted as biologically plausible mechanisms for adjusting unit outputs in neural networks, such as tonic inputs or activation thresholds, we investigate the expressivity of neural networks with random weights where only biases are optimized. We provide theoretical and numerical evidence demonstrating that feedforward neural networks with fixed random weights can be trained to perform multiple tasks by learning biases only. We further show that an equivalent result holds for recurrent neural networks predicting dynamical system trajectories. Our results are relevant to neuroscience, where they demonstrate the potential for behaviourally relevant changes in dynamics without modifying synaptic weights, as well as for AI, where they shed light on multi-task methods such as bias fine-tuning and unit masking.

Modulation of metastable ensemble dynamics explains optimal coding at moderate arousal in auditory cortex.

Performance during perceptual decision-making exhibits an inverted-U relationship with arousal, but the underlying network mechanisms remain unclear. Here, we recorded from auditory cortex (A1) of behaving mice during passive tone presentation, while tracking arousal via pupillometry. We found that tone discriminability in A1 ensembles was optimal at intermediate arousal, revealing a population-level neural correlate of the inverted-U relationship. We explained this arousal-dependent coding using a spiking network model with a clustered architecture. Specifically, we show that optimal stimulus discriminability is achieved near a transition between a multi-attractor phase with metastable cluster dynamics (low arousal) and a single-attractor phase (high arousal). Additional signatures of this transition include arousal-induced reductions of overall neural variability and the extent of stimulus-induced variability quenching, which we observed in the empirical data. Our results elucidate computational principles underlying interactions between pupil-linked arousal, sensory processing, and neural variability, and suggest a role for phase transitions in explaining nonlinear modulations of cortical computations.

Probabilistic modeling reveals coordinated social interaction states and their multisensory bases.

Social behavior across animal species ranges from simple pairwise interactions to thousands of individuals coordinating goal-directed movements. Regardless of the scale, these interactions are governed by the interplay between multimodal sensory information and the internal state of each animal. Here, we investigate how animals use multiple sensory modalities to guide social behavior in the highly social zebrafish (Danio rerio) and uncover the complex features of pairwise interactions early in development. To identify distinct behaviors and understand how they vary over time, we developed a new hidden Markov model with constrained linear-model emissions to automatically classify states of coordinated interaction, using the movements of one animal to predict those of another. We discovered that social behaviors alternate between two interaction states within a single experimental session, distinguished by unique movements and timescales. Long-range interactions, akin to shoaling, rely on vision, while mechanosensation underlies rapid synchronized movements and parallel swimming, precursors of schooling. Altogether, we observe spontaneous interactions in pairs of fish, develop novel hidden Markov modeling to reveal two fundamental interaction modes, and identify the sensory systems involved in each. Our modeling approach to pairwise social interactions has broad applicability to a wide variety of naturalistic behaviors and species and solves the challenge of detecting transient couplings between quasi-periodic time series.

A reservoir of timescales emerges in recurrent circuits with heterogeneous neural assemblies.

The temporal activity of many physical and biological systems, from complex networks to neural circuits, exhibits fluctuations simultaneously varying over a large range of timescales. Long-tailed distributions of intrinsic timescales have been observed across neurons simultaneously recorded within the same cortical circuit. The mechanisms leading to this striking temporal heterogeneity are yet unknown. Here, we show that neural circuits, endowed with heterogeneous neural assemblies of different sizes, naturally generate multiple timescales of activity spanning several orders of magnitude. We develop an analytical theory using rate networks, supported by simulations of spiking networks with cell-type specific connectivity, to explain how neural timescales depend on assembly size and show that our model can naturally explain the long-tailed timescale distribution observed in the awake primate cortex. When driving recurrent networks of heterogeneous neural assemblies by a time-dependent broadband input, we found that large and small assemblies preferentially entrain slow and fast spectral components of the input, respectively. Our results suggest that heterogeneous assemblies can provide a biologically plausible mechanism for neural circuits to demix complex temporal input signals by transforming temporal into spatial neural codes via frequency-selective neural assemblies.

Decision-making dynamics are predicted by arousal and uninstructed movements.

During sensory-guided behavior, an animal's decision-making dynamics unfold through sequences of distinct performance states, even while stimulus-reward contingencies remain static. Little is known about the factors that underlie these changes in task performance. We hypothesize that these decision-making dynamics can be predicted by externally observable measures, such as uninstructed movements and changes in arousal. Here, combining behavioral experiments in mice with computational modeling, we uncovered lawful relationships between transitions in strategic task performance states and an animal's arousal and uninstructed movements. Using hidden Markov models applied to behavioral choices during sensory discrimination tasks, we found that animals fluctuate between minutes-long optimal, sub-optimal and disengaged performance states. Optimal state epochs were predicted by intermediate levels, and reduced variability, of pupil diameter, along with reduced variability in face movements and locomotion. Our results demonstrate that externally observable uninstructed behaviors can predict optimal performance states, and suggest mice regulate their arousal during optimal performance.

A reservoir of foraging decision variables in the mouse brain.

In any given situation, the environment can be parsed in different ways to yield decision variables (DVs) defining strategies useful for different tasks. It is generally presumed that the brain only computes a single DV defining the current behavioral strategy. Here to test this assumption, we recorded neural ensembles in the frontal cortex of mice performing a foraging task admitting multiple DVs. Methods developed to uncover the currently employed DV revealed the use of multiple strategies and occasional switches in strategy within sessions. Optogenetic manipulations showed that the secondary motor cortex (M2) is needed for mice to use the different DVs in the task. Surprisingly, we found that regardless of which DV best explained the current behavior, M2 activity concurrently encoded a full basis set of computations defining a reservoir of DVs appropriate for alternative tasks. This form of neural multiplexing may confer considerable advantages for learning and adaptive behavior.

Differential encoding of temporal context and expectation under representational drift across hierarchically connected areas.

The classic view that neural populations in sensory cortices preferentially encode responses to incoming stimuli has been strongly challenged by recent experimental studies. Despite the fact that a large fraction of variance of visual responses in rodents can be attributed to behavioral state and movements, trial-history, and salience, the effects of contextual modulations and expectations on sensory-evoked responses in visual and association areas remain elusive. Here, we present a comprehensive experimental and theoretical study showing that hierarchically connected visual and association areas differentially encode the temporal context and expectation of naturalistic visual stimuli, consistent with the theory of hierarchical predictive coding. We measured neural responses to expected and unexpected sequences of natural scenes in the primary visual cortex (V1), the posterior medial higher order visual area (PM), and retrosplenial cortex (RSP) using 2-photon imaging in behaving mice collected through the Allen Institute Mindscope's OpenScope program. We found that information about image identity in neural population activity depended on the temporal context of transitions preceding each scene, and decreased along the hierarchy. Furthermore, our analyses revealed that the conjunctive encoding of temporal context and image identity was modulated by expectations of sequential events. In V1 and PM, we found enhanced and specific responses to unexpected oddball images, signaling stimulus-specific expectation violation. In contrast, in RSP the population response to oddball presentation recapitulated the missing expected image rather than the oddball image. These differential responses along the hierarchy are consistent with classic theories of hierarchical predictive coding whereby higher areas encode predictions and lower areas encode deviations from expectation. We further found evidence for drift in visual responses on the timescale of minutes. Although activity drift was present in all areas, population responses in V1 and PM, but not in RSP, maintained stable encoding of visual information and representational geometry. Instead we found that RSP drift was independent of stimulus information, suggesting a role in generating an internal model of the environment in the temporal domain. Overall, our results establish temporal context and expectation as substantial encoding dimensions in the visual cortex subject to fast representational drift and suggest that hierarchically connected areas instantiate a predictive coding mechanism.

Neural mechanisms underlying the temporal organization of naturalistic animal behavior.

Naturalistic animal behavior exhibits a strikingly complex organization in the temporal domain, with variability arising from at least three sources: hierarchical, contextual, and stochastic. What neural mechanisms and computational principles underlie such intricate temporal features? In this review, we provide a critical assessment of the existing behavioral and neurophysiological evidence for these sources of temporal variability in naturalistic behavior. Recent research converges on an emergent mechanistic theory of temporal variability based on attractor neural networks and metastable dynamics, arising via coordinated interactions between mesoscopic neural circuits. We highlight the crucial role played by structural heterogeneities as well as noise from mesoscopic feedback loops in regulating flexible behavior. We assess the shortcomings and missing links in the current theoretical and experimental literature and propose new directions of investigation to fill these gaps.

Metastable attractors explain the variable timing of stable behavioral action sequences.

The timing of self-initiated actions shows large variability even when they are executed in stable, well-learned sequences. Could this mix of reliability and stochasticity arise within the same neural circuit? We trained rats to perform a stereotyped sequence of self-initiated actions and recorded neural ensemble activity in secondary motor cortex (M2), which is known to reflect trial-by-trial action-timing fluctuations. Using hidden Markov models, we established a dictionary between activity patterns and actions. We then showed that metastable attractors, representing activity patterns with a reliable sequential structure and large transition timing variability, could be produced by reciprocally coupling a high-dimensional recurrent network and a low-dimensional feedforward one. Transitions between attractors relied on correlated variability in this mesoscale feedback loop, predicting a specific structure of low-dimensional correlations that were empirically verified in M2 recordings. Our results suggest a novel mesoscale network motif based on correlated variability supporting naturalistic animal behavior.

Distance estimation from monocular cues in an ethological visuomotor task.

In natural contexts, sensory processing and motor output are closely coupled, which is reflected in the fact that many brain areas contain both sensory and movement signals. However, standard reductionist paradigms decouple sensory decisions from their natural motor consequences, and head-fixation prevents the natural sensory consequences of self-motion. In particular, movement through the environment provides a number of depth cues beyond stereo vision that are poorly understood. To study the integration of visual processing and motor output in a naturalistic task, we investigated distance estimation in freely moving mice. We found that mice use vision to accurately jump across a variable gap, thus directly coupling a visual computation to its corresponding ethological motor output. Monocular eyelid suture did not affect gap jumping success, thus mice can use cues that do not depend on binocular disparity and stereo vision. Under monocular conditions, mice altered their head positioning and performed more vertical head movements, consistent with a shift from using stereopsis to other monocular cues, such as motion or position parallax. Finally, optogenetic suppression of primary visual cortex impaired task performance under both binocular and monocular conditions when optical fiber placement was localized to binocular or monocular zone V1, respectively. Together, these results show that mice can use monocular cues, relying on visual cortex, to accurately judge distance. Furthermore, this behavioral paradigm provides a foundation for studying how neural circuits convert sensory information into ethological motor output.

Estimating the Unique Information of Continuous Variables.

The integration and transfer of information from multiple sources to multiple targets is a core motive of neural systems. The emerging field of partial information decomposition (PID) provides a novel information-theoretic lens into these mechanisms by identifying synergistic, redundant, and unique contributions to the mutual information between one and several variables. While many works have studied aspects of PID for Gaussian and discrete distributions, the case of general continuous distributions is still uncharted territory. In this work we present a method for estimating the unique information in continuous distributions, for the case of one versus two variables. Our method solves the associated optimization problem over the space of distributions with fixed bivariate marginals by combining copula decompositions and techniques developed to optimize variational autoencoders. We obtain excellent agreement with known analytic results for Gaussians, and illustrate the power of our new approach in several brain-inspired neural models. Our method is capable of recovering the effective connectivity of a chaotic network of rate neurons, and uncovers a complex trade-off between redundancy, synergy and unique information in recurrent networks trained to solve a generalized XOR task.

Hippocampal Network Reorganization Underlies the Formation of a Temporal Association Memory.

Episodic memory requires linking events in time, a function dependent on the hippocampus. In "trace" fear conditioning, animals learn to associate a neutral cue with an aversive stimulus despite their separation in time by a delay period on the order of tens of seconds. But how this temporal association forms remains unclear. Here we use two-photon calcium imaging of neural population dynamics throughout the course of learning and show that, in contrast to previous theories, hippocampal CA1 does not generate persistent activity to bridge the delay. Instead, learning is concomitant with broad changes in the active neural population. Although neural responses were stochastic in time, cue identity could be read out from population activity over longer timescales after learning. These results question the ubiquity of seconds-long neural sequences during temporal association learning and suggest that trace fear conditioning relies on mechanisms that differ from persistent activity accounts of working memory.

Cortical computations via metastable activity.

Metastable brain dynamics are characterized by abrupt, jump-like modulations so that the neural activity in single trials appears to unfold as a sequence of discrete, quasi-stationary 'states'. Evidence that cortical neural activity unfolds as a sequence of metastable states is accumulating at fast pace. Metastable activity occurs both in response to an external stimulus and during ongoing, self-generated activity. These spontaneous metastable states are increasingly found to subserve internal representations that are not locked to external triggers, including states of deliberations, attention and expectation. Moreover, decoding stimuli or decisions via metastable states can be carried out trial-by-trial. Focusing on metastability will allow us to shift our perspective on neural coding from traditional concepts based on trial-averaging to models based on dynamic ensemble representations. Recent theoretical work has started to characterize the mechanistic origin and potential roles of metastable representations. In this article we review recent findings on metastable activity, how it may arise in biologically realistic models, and its potential role for representing internal states as well as relevant task variables.