Naturalistic animal behavior exhibits a striking amount of variability in the temporal domain (Figure 1A). An individual animal’s behavioral variability can be decomposed across at least three axes: hierarchical, contextual, and stochastic. The first source of variability originates from the vast hierarchy of timescales underlying self-initiated, spontaneous behavior ranging from milliseconds to minutes in animals (and to years for humans). On the sub-second timescale, animals perform fast movements varying from tens to hundreds of milliseconds. In rodents, these movements include whisking, sniffing, and moving their limbs. On the slower timescales of seconds, animals concatenate these fast movements into behavioral sequences of self-initiated actions, such as exploratory sequences (moving around an object while sniffing, whisking, and wagging their noses) or locomotion sequences (coordinating limb and head movements to reach a landmark). These sequences follow specific syntax rules (Berridge et al., 1987) and can last several seconds. On the timescales of minutes or longer, mice may repeat the ‘walk and explore’ behavioral sequence multiple times, when engaged in some specific activity, such as foraging, persisting toward their goal for long periods. In these simple examples, a freely moving mouse exhibits behavior whose temporal organization varies over several orders of magnitudes simultaneously, ranging from the sub-second scale (actions), to several seconds (behavioral sequences), to minutes (goals to attain). A leading theory to explain the temporal organization of naturalistic behavior is that behavioral action sequences arrange in a hierarchical structure (Tinbergen, 1951; Dawkins, 1976; Simon, 1962), where actions are nested into behavioral sequences that are then grouped into activities. This hierarchy of timescales is ubiquitously observed across species during naturalistic behavior.

Figure 1

Download asset Open asset

The second source of temporal variability stems from the different contexts in which a specific behavior can be performed. For example, reaction times to a delivered stimulus can be consistently faster when an animals is expecting it compared to slower reaction times when the stimulus is unexpected (Jaramillo and Zador, 2011). Contextual sources can be either internally generated (modulations of brain state, arousal, expectation) or externally driven (pharmacological, optogenetic or genetic manipulations; or changes in task difficulty or environment).

After controlling for all known contextual effects, an individual’s behavior still exhibits a large amount of residual temporal variability across repetitions of the same behavioral unit. This residual variability across repetitions can be quantified as a property of the distribution of movement durations, which is typically very skewed with an exponential tail (Wiltschko et al., 2015; Findley et al., 2021; Recanatesi et al., 2022), suggesting a stochastic origin. We define as stochastic variability the fraction of variability in the expression of a particular behavioral unit, which cannot be explained by any readily measurable variables.

These three aspects of temporal variability (hierarchical, contextual, stochastic) can all be observed in the naturalistic behavior of a single individual during their lifetime. Other sources of behavioral variability have also been investigated. Variability in movement execution and body posture (Dhawale et al., 2017) may originate from the motor periphery, such as noise in force production within muscles (van Beers et al., 2004), or from the deterministic chaotic dynamics arising from neural activity (Costa et al., 2021) or the biomechanical structure of an animal’s body (Loveless et al., 2019). One additional source is individual phenotypic variability across different animals, an important aspect of behavior in ethological and evolutionary light (Honegger and de Bivort, 2018). Although we will briefly discuss some features of variability in movement execution and phenotypic individuality, the main focus of this review will be on the three main axes of variability described above, which pertain to the behavior of a single animal.

From this overview, we conclude that the temporal structure of naturalistic, self-initiated behavior can be decomposed along at least three axes of temporal variability (Figure 1A): a large hierarchy of simultaneous timescales at which behavior unfolds, ranging from milliseconds to minutes; contextual modulations affecting the expression of behavior at each level of this hierarchy; and stochasticity in the variability of the same behavioral unit across repetition in the same context. These three axes may or may not interact depending on the scenarios and the species. Here, we will review recent theoretical and computational results establishing the foundations of a mechanistic theory that explains how these three sources of temporal variability can arise from biologically plausible computational motifs. More specifically, we will address the following questions:

1. Neural representation: How are self-initiated actions represented in the brain and how are they concatenated into behavioral sequences?

2. Stochasticity: What neural mechanisms underlie the temporal variability observed in behavioral units across multiple repetitions within the same context?

3. Context: How do contextual modulations affect the temporal variability in behavior, enabling flexibility in action timing and behavioral sequence structure?

4. Hierarchy: How do neural circuits generate the vast hierarchy of timescales from milliseconds to minutes, hallmark of naturalistic behavior?

The mechanistic approach we will review is based on the theory of metastable attractors (Figure 1B), which is emerging as a unifying principle expounding many different aspects of the dynamics and function of neural circuits (La Camera et al., 2019). We will first establish a precise correspondence between behavioral units and neural attractors at the level of self-initiated actions (the lowest level of the temporal hierarchy). Then we will show how the emergence of behavioral sequences originates from sequences of metastable attractors. Our starting point is the observation that transitions between metastable attractors can be driven by the neural variability internally generated within a local recurrent circuit. This mechanism can naturally explain the action timing variability of stochastic origin. We will examine biologically plausible mesoscopic circuits that can learn to flexibly execute complex behavioral sequences. We will then review the neural mechanisms underlying contextual modulations of behavioral variability. We will show that the average transition time between metastable attractors can be regulated by changes in single-cell gain. Gain modulation is a principled neural mechanism mediating the effects of context, which can be induced by either internal or external perturbations and supported by different neuromodulatory and cortico-cortical pathways; or by external pharmacological or experimental interventions. Finally, we will review how a large hierarchy of timescales can naturally and robustly emerge from heterogeneities in a circuit’s structural connectivity motifs, such as neural clusters with heterogeneous sizes. Although most of the review is focused on the behavior of individual animals, we discuss how recent results on multianimal interactions and social behavior may challenge existing theories of naturalistic behavior and brain function. Appendices 1–3 provide guides to computational methods for behavioral video analyses and modeling; theoretical and experimental aspects of attractors dynamics; and biologically plausible models of metastable attractors.

The stochastic nature of naturalistic behavior

The study of naturalistic behavior based on animal videos has recently undergone a revolution due to the spectacular accuracy and efficiency of computational methods for animal pose tracking (Stephens et al., 2008; Berman et al., 2014; Hong et al., 2015; Mathis et al., 2018; Pereira et al., 2019; Nourizonoz et al., 2020; Segalin et al., 2020); see Appendix 1 and Table 1 for details). These new methods work across species and conditions, ushering a new era for computational neuroethology (Anderson and Perona, 2014; Brown and de Bivort, 2018; Datta et al., 2019). They have led to uncovering a quantitative classification of self-initiated behavior revealing a stunning amount of variability both in its lexical features (which actions to choose, in which order) and in its temporal dimension (when to act) (Berman et al., 2016; Wiltschko et al., 2015; Markowitz et al., 2018; Marshall et al., 2021; Recanatesi et al., 2022). Different ways to characterize the behavioral repertoire on short timescales have been developed, depending on the underlying assumptions of whether the basic units of behavior are discrete or continuous (Appendix 1). One can define stereotypical behaviors or postures by clustering probability density maps of spectrotemporal features extracted from behavioral videos, as in the example of fruit flies (Berman et al., 2014; Berman et al., 2016; (Figure 2A) or rats (Marshall et al., 2021; Figure 7B). Alternatively, one can capture actions or postures as discrete states of a state–space model based on Markovian dynamics, each state represented as a latent state autoregressive trajectory accounting for stochastic movement variability (Figure 2). In the case of discretization of the behavioral repertoire, the definition of the smallest building block of behavior depends on the level of granularity that best suits the scientific question being investigated. Within this discrete framework, state–space models were applied successfully to Caenorhabditis elegans (Linderman et al., 2019), Drosophila (Tao et al., 2019), zebrafish (Johnson et al., 2020), and rodents (Wiltschko et al., 2015; Markowitz et al., 2018). At the timescale of actions and postures, transition times between consecutive actions are well described by a Poisson process (Killeen and Fetterman, 1988), characterized by a right-skewed distribution of inter-action intervals (Figure 3). In this and the next section, we will focus on short timescales (up to a few seconds) where state–space models can provide parsimonious accounts of behavior. However, these models fail to account for longer timescale and non-Markovian structure in behavior, and in later sections we will move to a more data-driven approach to investigate these aspects of behavior (see Appendix 1).

Table 1

Action	The simplest building blocks of behavior are at the lowest level of the hierarchy. This definition depends on the level of granularity (spatiotemporal resolution) required and the scientific questions to be investigated (see Appendix 1). A widely used definition of action is a short stereotypical trajectory in posture space (synonyms include movemes, syllables [Anderson and Perona, 2014; Brown and de Bivort, 2018]), operationally defined as a discrete latent trajectory of an autoregressive state–space model fit to pose-tracking time-series data (Figure 2; Wiltschko et al., 2015; Findley et al., 2021). An alternative definition is in terms of short spectrotemporal representations from a time-frequency analysis of videos (Berman et al., 2016; Marshall et al., 2021). Examples include poking in or poking out of a nose port, waiting at a port, pressing a lever.
Behavioral action sequence	A combination of actions concatenated in a meaningful yet stereotyped way, lasting up to a few seconds. A sequence can occur during trial-based experimental protocols (e.g., a short sequence of actions aimed at obtaining a reward in an operant conditioning task [Geddes et al., 2018; Murakami et al., 2014]; running between opposite ends of a linear track [Maboudi et al., 2018]) or during spontaneous periods (e.g., repeatedly scratching own head; picking up and manipulating an object).
Activity	A concatenation of multiple behavioral sequences, often repeated and of variable duration, typically aimed at obtaining a goal and lasting up to minutes or even hours. Examples include grooming, foraging, mating, feeding, and exploration. Activities typically unfold in naturalistic freely moving settings devoid of experimenter-controlled trial structure.

Figure 2

Download asset Open asset

Figure 3

Download asset Open asset

The second source of temporal variability in naturalistic behavior arises from contextual modulations, which can be internally or externally driven. When internally driven, they may arise from changes in brain or behavioral state such as arousal, expectation, or task engagement. When externally driven, they may arise from changes in environmental variables, from the experimenter’s imposed task conditions, or from manipulations such as pharmacological, optogenetic, or genetic ones. Contextual modulations can affect several qualitatively different aspects of behavioral units at each level of the hierarchy (actions, sequences, activities): average duration; usage frequency; and transition probabilities between units. Moreover, context may also change the motor execution of a behavioral unit, for example, by improving the vigor of a certain movement upon learning or motivation. For each type of modulation, we will give several examples and review computational mechanisms that may explain them.

Action timing

The distribution of self-initiated action durations typically exhibits large variability, whose characteristic timescale can be extracted from their average duration (Figure 3). The average timing of an action is strongly modulated by contextual factors, both internally and externally driven. Examples of internally generated contextual factors include expectation and history effects. When events occur at predictable instants, anticipation improves performance such as reaction times. This classic effect of expectation was documented in an auditory two-alternative choice task (Jaramillo and Zador, 2011), where freely moving rats were rewarded for correctly discriminating the carrier frequency of a frequency-modulated target sound immersed in pure-tone distractors (Figure 4A). The target could occur early or late within each sound presentation, and temporal expectations on target timing were modulated by changing the ratio of trials with early or late targets within each block. When manipulating the expectations about sound timing, valid expectations accelerated reaction times and improved detection accuracy, showing enhanced perception. The auditory cortex is necessary to perform this task, and firing rates in auditory cortex populations are modulated by temporal expectations.

Figure 4

Download asset Open asset

Contextual changes in brain state may also be induced by varying levels of neuromodulators. In the self-initiated waiting task of Figure 3A, optogenetic activation of serotonergic neurons in the dorsal raphe nucleus selectively prolonged the waiting period, leading to a more patient and less impulsive behavior, but did not affect the timing of other self-initiated actions (Fonseca et al., 2015). Internally generated contextual modulations include history effects, which can affect the timing of self-initiated actions. In an operant conditioning task, where freely moving mice learned to press a lever for a minimum duration to earn a reward (Fan et al., 2012), the distribution of action timing showed dependence on the outcome of the preceding trial. After a rewarded trial, mice exhibited longer latency to initiate the next trial, but shorter press durations; after a failure, the opposite behavior occurred, with shorter latency to engage and longer press durations. Trial history effects are complex and action-specific and depend on several other factors, including prior movements, and wane with increasing inter-trial intervals. Inactivation experiments showed that these effects rely on frontal areas such as medial prefrontal and secondary motor cortices (Murakami et al., 2014; Murakami et al., 2017; Schreiner et al., 2021).

Although the contextual modulations considered so far occur on a fast timescale of a few trials, they may also be the consequence of associative learning. In a lever press task, mice learned to adjust the average duration of the lever press to different criteria in three different conditions where lever presses were always rewarded regardless of duration or only rewarded if longer than 800 or 1600 ms. Within each of the three conditions, the distribution of action timing exhibited large temporal variability, yet the average duration was starkly different between the three criteria as the mice learned the different criteria (Fan et al., 2012; Schreiner et al., 2021). Reaction times to sensory stimuli and self-initiated waiting behavior, in the form of long lever presses or nose-pokes, have emerged as a fruitful approach to test hypotheses on contextual modulations and decision-making in naturalistic scenarios.

Behavioral sequences

Contextual modulations of temporal variability may affect other aspects of natural behavior beyond average action timing, such as the frequency of occurrence of an action (‘state usage’) or the transition probabilities between actions within a behavioral sequence (Figure 5A). These effects can be uncovered by analyzing freely moving behavioral videos with state–space models, such as the autoregressive hidden Markov model (HMM) (Wiltschko et al., 2015; Figures 2B and 5A and Appendix 1). Differences in state usage and transition probabilities between conditions may shed light on the computational strategies animals deploy to solve complex ethological tasks, such as those involving sensorimotor integration. For example, in a distance estimation task where mice were trained to jump across a variable gap, a comparison of monocular and binocular mice revealed the different visually guided strategies mice may use to perform a successful jump. Mice performed more vertical head movements under monocular conditions compared to control (Figure 5A, states 2 and 3 occur more frequently in the monocular condition), revealing a reliance on motion parallax cues (Parker et al., 2021).

Figure 5

Download asset Open asset

During ongoing periods, in the absence of a task, animal behavior features a large variety of actions and behavioral sequences (Figure 2). Experimentally controlled manipulations can lead to strong changes in ongoing behavior reflected both in changes of state usage and of transition probabilities between actions, resulting in different repertoires of behavioral sequences. Examples of manipulations include exposing mice to innately aversive odors and other changes in their surrounding environment; optogenetic activation of corticostriatal pathways (Wiltschko et al., 2015); and pharmacological treatment (Wiltschko et al., 2020). In the latter study, a classification analysis predicted with high accuracy which drug and specific dose was administered to the mice from a large panel of compounds at multiple doses. Comparison of state usage and transition rates can also reveal subtle phenotypical changes in the structure of ongoing behavior in genetically modified mice compared to wildtype ones. This phenotypic fingerprinting has led to insights into the behavior of mouse models of autism spectrum disorder (Wiltschko et al., 2015; Wiltschko et al., 2020; Klibaite et al., 2021).

The temporal organization of naturalistic behavior exhibits a clear hierarchical structure (Tinbergen, 1951; Dawkins, 1976; Simon, 1962), where actions are nested into behavioral sequences that are then grouped into activities. Higher levels in the hierarchy emerge at longer timescales: actions/movements occur on a sub-second scale, behavioral sequences span at most a few seconds and activities last for longer periods of several seconds to minutes. The crucial aspect of this behavioral hierarchy is its complexity: an animal’s behavior unfolds along all timescales simultaneously. What is the organization of this vast spatiotemporal hierarchy? What are the neural mechanisms supporting and generating this nested temporal structure?

A case study: C. elegans locomotor behavior

Remarkable progress in both behavioral and neurophysiological aspects of the hierarchy has been made in the worm C. elegans (Figure 6). The worm locomotor behavior exhibits a clear hierarchical organization with three distinct timescales: sub-second body-bends and head-casts, second-long crawling undulations, and slow reverse-forward cycles (Gomez-Marin et al., 2016). At the neuronal level, this hierarchy is generated by nested neuronal dynamics where upper-level motor programs are supported by slow activity spread across many neurons, while lower-level behaviors are represented by fast local dynamics in small multifunctional populations (Kaplan et al., 2020). Persistent activity driving higher-level behaviors gates faster activity driving lower-level behaviors, such that, at lower levels, neurons show dynamics spanning multiple timescales simultaneously (Hallinen et al., 2021). Specific lower-level behaviors may only be accessed via switches at upper levels, generating a nonoverlapping, tree-like hierarchy, in which no lower-level state is connected to multiple upper-level states.

Figure 6

Download asset Open asset

At the top of this motor hierarchy, we find a much longer-lasting organization of states in terms of exploration, exploitation, and quiescence. In contrast to the strict, tree-like structure observed in the motor hierarchy, lower-level motor features are shared across these states, albeit with different frequency of occurrence. Whereas the motor hierarchy is directly generated by neuronal activity, this state-level hierarchy may rely on neuromodulation (Ben Arous et al., 2009; Flavell et al., 2013)

Behavioral hierarchies in flies and rodents

How much of this tight correspondence between behavioral and neural hierarchies generalizes from worms to insects and mammals? Are action sequences organized as a chain? Alternatively, is there a hierarchical structure where individual actions, intermediate subsequences, and overall sequences can be flexibly combined? A chain-like organization would require a single controller concatenating actions, but could be prone to error or disruption. A hierarchical structure could be error-tolerant and flexible at the cost of requiring controllers at different levels of the hierarchy (Geddes et al., 2018). We will first examine compelling evidence from the fruit fly and then discuss emerging results in rodent experiments.

By analyzing videos of ground-based fruit fly during long sessions of spontaneous behavior (Figure 2B), Berman et al., 2014 and Berman et al., 2016 mapped the entire behavioral space of freely moving flies, identifying a hundred stereotyped, frequently reoccurring actions, interspersed with bouts of nonstereotyped behaviors. Recurring behavioral categories emerged as peaks in the behavioral space probability landscape, labeled as walking, running, head grooming, and wing grooming (Figure 7A). In order to uncover the organization of behavior at different timescales, the authors estimated which behavioral representations (movements, sequences, or activities) could optimally predict the fly’s future behavior on different temporal horizons (from 50 ms to minutes), applying the information bottleneck method (Tishby et al., 2000). This predictive algorithm revealed multiple timescales in the fly behavior, organized into a hierarchical structure reflecting the underlying behavioral programs and internal states. The near future could be optimally predicted by segmenting behavior according to actions at the fastest level of the hierarchy. The optimal representation of behavior that could optimally predict the distant future up to minutes away was based on slower, coarser groups of actions grouped into activities. These longer timescales manifest as nested blocks in the transition probability matrix, implying that a strongly non-Markovian structure emerges on longer timescales (Alba et al., 2020; Figure 7A).

Figure 7

Download asset Open asset

A hallmark of the fly behavior emerging from this analysis is that the different branches of this hierarchy are segregated into a tree-like, nonoverlapping structure: for example, actions occurring during grooming do not occur during locomotion. This multiscale representation was leveraged to dissect the descending motor pathways in the fly. Optogenetic activation of single neurons during spontaneous behavior revealed that most of the descending neurons drove stereotyped behaviors that where shared by multiple neurons and often depended on the behavioral state prior to activation (Cande et al., 2018). An alternative statistical approach based on a hierarchical HMM revealed that, although all flies use the same set of low-level locomotor features, individual flies vary considerably in the high-level temporal structure of locomotion, and in how this usage is modulated by different odors (Tao et al., 2019). This behavioral idiosyncrasy of individual-to-individual phenotypic variability has been traced back to specific genes (Ayroles et al., 2014) regulating neural activity (Buchanan et al., 2015) in the central complex of the fly brain. This series of studies suggest that the fly behavior is organized hierarchically in a tree-like structure from actions to sequences to activities. This hierarchy seems to lack flexibility, such that behavioral units at lower levels are segregated into different hierarchical branches and cannot be rearranged at higher levels. Although a more flexible structure seems to emerge during social interactions (see above and Calhoun et al., 2019).

What is the structure of the behavioral hierarchy in mammals? We will start by considering the first two levels of the hierarchy, namely, how actions concatenate into sequences. A recent experimental tour de force demonstrated the existence of a hierarchical structure in the learning and execution of heterogeneous sequences in a freely moving operant task (Geddes et al., 2018). Mice learned to perform a ‘penguin dance’ consisting of a sequence of two or three consecutive left lever presses (LL or LLL) followed by a sequence of two or more right lever presses (RR or RRR) to obtain a reward. Mice acquired the sequence hierarchically rather than sequentially, a learning scheme that is inconsistent with the classic reinforcement learning paradigm, which predicts sequence learning occurs in the reverse order of execution (Sutton and Barto, 2018). Using closed-loop optogenetic stimulation, the authors revealed the differential roles played by striatal direct and indirect pathways in controlling, respectively, the expression of a single action (either L or R), or a fast switch from one subsequence to the next (from the LL block to RR, and from RR to the reward approach).

What is the structure of the rodent behavioral hierarchy at slower timescales? A multiscale analysis of ongoing behavior was performed using rat body piercing, which allowed for tracking of the three-dimensional pose with high accuracy (Figure 7B; Marshall et al., 2021). A watershed algorithm applied to behavioral probability density maps revealed the emergence of hierarchical behavioral categories at different temporal scales. Examining the behavioral transition matrix at different timescales, signature of non-Markovian dynamics peaked at 10–100 s timescales. On 15 s timescales, pattern sequences emerged featuring sequentially ordered actions, such as grooming sequences of the face followed by the body. On minute-long timescales, states of varying arousal or task engagement emerged, lacking stereotyped sequential ordering. Although the results of Marshall et al., 2021 seem to suggest the presence of a flexible, overlapping structure in the rat hierarchy, consistent with Geddes et al., 2018 and different from the tree-like hierarchy in flies (Berman et al., 2016), the analysis of Marshall et al., 2021 lacked the information bottleneck step, which was crucial to capture the fly behavior on long timescales. Future work may shed light on the nature of behavioral hierarchies in mammals using the information bottleneck.

Manipulation experiments could help clarify the structure of the rodent behavioral hierarchy and whether it allows for flexible behavioral sequences. Earlier studies of contextual modulations of rodent freely moving behavior by pharmacological or genetic manipulations were mostly confined to comparison of usage statistics of single actions (Wiltschko et al., 2015). However, more recent studies (Marshall et al., 2021; Klibaite et al., 2021), which mapped the rodent behavioral hierarchy, allowed a multiscale assessment of contextual effects. In a rat model of the fragile X syndrome, while mutant and wildtype rats had similar locomotor behavior, the former showed abnormally long grooming epochs, characterized by different behavioral sequences and states compared to wildtypes (Figure 7C). This study highlights the advantages of multiscale comparative taxonomy of naturalistic behavior to investigate behavioral manifestations of complex conditions such as autism spectrum disorder and classify the behavioral effects of different drugs. Another benefit of this multiscale approach is to allow the investigation of the individuality of rodent behavior, revealing that, although all animals drew from a common set of temporal patterns, differences between individual rats were more pronounced at longer timescales compared to short ones (Marshall et al., 2021). It would be interesting to further explore the features of individuality across species as a way of testing ecological ideas such as ‘bet-hedging’ (Honegger and de Bivort, 2018).

Multiple timescales of neural activity

What are the neural mechanisms generating the hierarchy of timescales observed during naturalistic behavior? Is there evidence that neural activity is simultaneously varying over multiple timescales? Although no studies directly addressed these questions yet, a number of experimental and theoretical approaches have provided evidence for multiple timescales of neural activity. Some evidence for temporal heterogeneity in neural activity was reported in restrained animals during stereotyped behavioral assays. A heterogeneous distribution of timescales of neural activity was found in the brainstem of the restrained larval zebrafish by measuring the decay time constant of persistent firing across a population of neurons comprising the oculomotor velocity-to-position neural integrator (Figure 8A; Miri et al., 2011). The decay times varied over a vast range 0.5–50 s across cells in individual larvae. This heterogeneous distribution of timescales was later confirmed in the primate oculomotor brainstem (Joshua et al., 2013), suggesting that timescale heterogeneity is a common feature of oculomotor integrators conserved across phylogeny. Single-neuron activity may also encode a long memory of task-related variables. In head-fixed monkeys performing a competitive game task, temporal traces of the reward delivered in previous trials were encoded in single-neuron spiking activity in frontal areas over a wide range of timescales, obeying a power law distribution up to 10 consecutive trials (Bernacchia et al., 2011).

Figure 8

Download asset Open asset

The timescale of intrinsic fluctuations in spiking activity can be also estimated from single-neuron spike autocorrelation functions (Figure 8B). In awake head-fixed primates, during periods of ongoing activity, the distribution of intrinsic autocorrelation timescales within the same cortical circuit was found to be right-skewed and approximately lognormal (Figure 8B), in the range from 10 ms to 1 s (Cavanagh et al., 2016). Moreover, comparison of the population-averaged autocorrelations during ongoing periods revealed a hierarchical structure across cortical areas, varying from 50 to 350 ms along the occipital-to-frontal axis (Murray et al., 2014; Figure 8C).

While all these results were obtained in restrained animals, it is an open question whether neural activity during naturalistic behavior exhibits temporal hierarchies similar to those observed in behavior. Current evidence from freely moving rodents engaged in waiting tasks (Murakami et al., 2017; Schreiner et al., 2021; Recanatesi et al., 2022) revealed the presence of multiple timescales in single-neuron activity from secondary motor and prefrontal cortices. These timescales range from the sub-second scale (single attractors), to a few seconds (attractor sequences), to tens of seconds or minutes (trial history dependence). It is tantalizing to speculate that even longer timescales may be present for neural activity to be able to generate the vast hierarchy of timescales observed in naturalistic behavior (Figure 7). Evidence from associative learning tasks in rodents found that population activity in hippocampal CA1 and CA3 encodes multiplexed information about several aspects of the task occurring on multiple trials such as context, place, value, and objects (McKenzie et al., 2014). Although an explicit analysis of temporal correlations was not carried out in this study, these results suggest that a hierarchy of timescales may be present in the hippocampus and emerge during learning.

Future directions: Neural mechanisms generating temporal hierarchies

Future studies should address how the hierarchy of timescales found in behavior may emerge from computational mechanisms. There are several missing links along this path. First, although temporally heterogeneous neural activity was found in several brain areas and species under different experimental conditions, no study to date investigated the presence of temporal hierarchies in neural activity during naturalistic behavior. All the necessary tools are available: recent advances in neurotechnology demonstrate the feasibility of chronic recordings of large neural populations during freely moving behavior (Juavinett et al., 2019; van Daal et al., 2021); simultaneously performing behavioral classification analyses from pose tracking software (Mathis et al., 2018; Marshall et al., 2021; Pereira et al., 2019). This first piece of the puzzle is thus within reach.

On the theory side, although recurrent dynamics can generate heterogeneous distributions of timescales (Figure 8E), this model needs to be extended to a fully realistic framework for explaining nested temporal hierarchies in naturalistic behavior such as the proposed mesoscopic circuit in Figure 5C. Here, we propose a simple roadmap to bridge this gap, building on some of the theoretical ideas we reviewed above. This theory is based on a hierarchical structure that, for lack of a better term, we will denote ‘Attractors all the way down’ (Figure 9). We start from the observation that self-initiated actions within a behavioral sequence are represented as metastable attractors in secondary motor cortex (Inagaki et al., 2019; Recanatesi et al., 2022; Figure 3). We then extend this observation to a general principle positing that behavioral units at each level of the temporal hierarchy (not only actions, but also sequences, activities) are represented by metastable attractors. The stochasticity in behavioral unit duration can be achieved by generating transitions between attractors at each level via low-dimensional variability, arising from mesoscopic feedback loops involving cortex and subcortical areas (Recanatesi et al., 2022; Figure 3D). The average transition time between behavioral units at each level will then depend on the barrier height separating the corresponding attractors (Cao et al., 2016; Mazzucato et al., 2019; Wyrick and Mazzucato, 2021). As one moves up the hierarchy from actions (fast transitions = lower barriers) to sequences and activities (slow transitions = higher barriers), the potential wells separating the corresponding attractors become deeper and the basins of attraction wider. This increase in timescales can be achieved in a biological manner by assuming that increasingly larger neural populations encode for slower behavioral features (Stern et al., 2021; see Figure 8E), which can be realized in a biologically plausible way as a hierarchy of clusters within clusters (Schaub et al., 2015). Contextual modulations in average duration of a behavioral unit at each level of the hierarchy can be implemented by top-down changes of the barrier heights between attractors via gain modulation (Figure 4C). The new theoretical ingredient required for this theory to work is the presence of a nested structure in the attractor landscape, whereas the basin of attraction of lower-level attractors (actions) is contained within the basins of higher level attractors (sequences) all the way up to the largest basins representing long-term activities. The architecture of this model provides some immediate predictions for neural activity. First, neurons exhibit conjunctive selectivity to multiple variables at different levels of the hierarchy, which has been observed, for example, in rodent hippocampus during freely moving tasks (McKenzie et al., 2014) and in primate prefrontal cortex during complex decision-making tasks (Rigotti et al., 2013). Second, single-cell activity is modulated by many attractors, in agreement with the multistable activity observed in sensory (Mazzucato et al., 2015) and frontal areas (Recanatesi et al., 2022) where neuronal representations of attractors are dense rather than sparse. Third, fluctuations in the activity of single neurons should vary over multiple timescales simultaneously, encoding information about multiple levels of the hierarchy; this is the structure that was observed in C. elegans where neurons representing movements exhibit both fast and for slow fluctuations correlated to multiple levels of the behavioral hierarchy (Kaplan et al., 2020; Figure 6).

Figure 9

Download asset Open asset

Which network architectures can implement a hierarchical attractor landscape? The behavioral hierarchy discovered in flies (Berman et al., 2016) can be approximated as a tree: behavioral units at a lower level of the hierarchy only occur during a specific unit at a higher level of the hierarchy (Figure 7). For example, movements occurring during the fruit fly locomotion activities never occur during idle activities. This structure is typical of phylogenetic classification in taxonomy, where the distance from the root to the leaves is the same for every leaf and the tree is called ultrametric (Rammal et al., 1986). It is not known whether behavioral trees are ultrametric; in order to clarify this structure, a notion of distance in behavioral space will have to be introduced and examined. A classic model of hierarchical attractors, which could potentially capture the tree-based hierarchy, is an extension of the Hopfield network where the stored pattern is hierarchically arranged into a tree (Parga and Virasoro, 1986; Figure 9). This class of models was originally proposed to explain the effect of word semantic categorization during memory encoding, storage, and retrieval. A contamination between the domains of naturalistic behavior and natural language processing has been proposed early on (Dawkins, 1976) and recently applied to describe C. elegans (Gomez-Marin et al., 2016; Gupta and Gomez-Marin, 2019) and larval zebrafish behavior (Reddy et al., 2022). It is still an open question to determine the precise structure of the behavioral hierarchy in mammals. The results of Marshall et al., 2021 seem to suggest that the hierarchy in rats may not be tree-like as the one observed in flies (Berman et al., 2016). However, the analysis of the rodent behavior in Marshall et al., 2021 lacked the information bottleneck step, which was crucial to infer the structure of the fly hierarchy at long timescale. On general grounds, it would be interesting to investigate whether the mammalian hierarchical motor system allows for flexible and adaptable behaviors, by utilizing the same motor elements across different sequences in a hierarchical manner, as represented in the mesoscopic circuit models of Murray and Escola, 2017; Logiaco and Escola, 2020; Logiaco et al., 2021 (see Figure 5) and suggested by experimental evidence in Geddes et al., 2018. This flexible hierarchy could be realized by overlapping branches and would be different from the tree-like structure found in the fly.

Are the computational mechanisms generating the tree-like hierarchy sufficient to generate flexible hierarchies? Recent theoretical work highlighted the importance of a neural circuit architecture that segregates the motor preparation and execution in different areas. In the cortico–thalamic–basal ganglia model of Figure 5B, flexible sequences can be generated by rearranging existing motifs as well as by learning new motifs without interfering with previous ones (Logiaco et al., 2021). Moreover, flexible sequences can be learned using biologically plausible Hebbian plasticity in the striatum (Murray and Escola, 2017). These theoretical models provide an exciting blueprint for further investigation into learning and expression of flexible behavioral hierarchies.

What are potential neural implementations of this network architecture? In the case of mammals, at the lowest level of the hierarchy, neural populations in the rodent secondary motor cortex (M2) generate metastable attractors representing upcoming actions (Inagaki et al., 2019; Recanatesi et al., 2022). Other subcortical areas, such as the basal ganglia or the thalamus, are likely involved in the generation of complex behavioral sequences (Murray and Escola, 2017; Logiaco et al., 2021; Recanatesi et al., 2022). How are higher-order behavioral units, such as sequences and activities, encoded? One possibility is that they could also be encoded in M2 attractors as recent evidence suggests M2 populations encode for a wide range of behavioral variables (Cazettes et al., 2021). We hypothesize an architecture featuring nested assemblies where larger populations, whose activity varies over progressively slower timescales (Stern et al., 2021; Schaub et al., 2015), hierarchically encode for slower features of behavior (from actions to sequences to activities). Experimental evidence shows that single neurons in M2 are selective to multiple actions (Recanatesi et al., 2022), suggesting that mixed selectivity (Rigotti et al., 2013) could play an important role in generating the high-dimensional hierarchical attractor landscape necessary to capture the complexity of naturalistic behavior. An alternative architecture might involve a distributed cortical circuit where the neural representations of behavioral units at different levels of the hierarchy are encoded in multiple frontal cortical areas, such as action sequences in M2, and activities in prefrontal cortex, which is known to control behavior on longer timescales such as trial history effects (Murakami et al., 2014; Murakami et al., 2017; Schreiner et al., 2021). In both scenarios, we hypothesize that while behavioral units are encoded as cortical attractors, transitions between attractor rely on feedback loops involving cortico–subcortical circuits (Murray and Escola, 2017; Logiaco et al., 2021; Recanatesi et al., 2022).

Behavioral energy landscape

The structure of naturalistic behavior emerging on long timescales in the fruit fly is a hierarchical tree-like organization (Figure 7A), which is consistent with an underlying neural circuit architecture based on hierarchical basins of attractions (Figure 9). Can we derive a representation of behavior as a complex energy landscape directly from the behavioral data itself? A promising approach is given by the probability density map approach (Berman et al., 2016). One could define the log-probability of the density map as an energy potential, where the transition rates between behavioral features yield a probability flux along this potential landscape. It is tantalizing to speculate that using the combination of energy gradient and probability flux one could derive a data-driven nonlinear dynamical system describing behavior (Wang, 2015). Other approaches to infer a potential energy landscape directly from data have been successfully applied to spike trains (Genkin and Engel, 2020; Duncker et al., 2019), although in a regime where the energy landscape has only a few minima. Alternatively, in chemical kinetics or fluid flow, methods based on the transfer operator formalism have been applied to find effective free energy landscapes and metastable states from experimental data or simulations (Bowman et al., 2013), and have been recently applied to animal behavior (Costa et al., 2021).

Neural mechanisms of flexible behavioral hierarchies

The model of nested basins of attractions proposed in Figure 9 can explain a tree-like behavioral hierarchy, where behavioral units at lower levels (e.g., actions) are not shared by different units at a higher level (e.g., sequences), as observed in fruit flies (Berman et al., 2016). Although the tree-like hierarchical organization emerged from assays where individual animals were monitored in isolation, recent studies of social behavior seem to challenge this structure. A large variety of qualitatively new behaviors arise from social interactions, including fighting, mating, and others. While some of these behaviors involve specialized behavioral units, others involve simultaneous execution of multiple behavioral units leading to multitasking. For example, during courtship, a male fly can simultaneously approach a female fly (locomotion) and sing, two behaviors that would be mutually exclusive in the absence of a female fly. As a consequence, the simple tree-like hierarchical organization of behavior observed in isolated individuals (Figure 8) might break down during social interactions and lead to a flexible organization where actions are shared between multiple sequences and activities.

1. 1. Ahamed T
   2. Costa AC
   3. Stephens GJ

   (2021) Capturing the continuous complexity of behaviour in Caenorhabditis elegans

   Nature Physics 17:275–283.

   https://doi.org/10.1038/s41567-020-01036-8

   • Google Scholar
2. Preprint

   1. Alba V
   2. Berman GJ
   3. Bialek W
   4. Shaevitz JW

   (2020) Exploring a Strongly Non-Markovian Animal Behavior

   ArXiv Preprint.

   https://arxiv.org/abs/2012.15681

   • Google Scholar
3. 1. Aljadeff J
   2. Stern M
   3. Sharpee T

   (2015) Transition to chaos in random networks with cell-type-specific connectivity

   Physical Review Letters 114:088101.

   https://doi.org/10.1103/PhysRevLett.114.088101

   • PubMed
   • Google Scholar
4. 1. Almaas E
   2. Kovács B
   3. Vicsek T
   4. Oltvai ZN
   5. Barabási AL

   (2004) Global organization of metabolic fluxes in the bacterium Escherichia coli

   Nature 427:839–843.

   https://doi.org/10.1038/nature02289

   • PubMed
   • Google Scholar
5. 1. Amit DJ
   2. Gutfreund H
   3. Sompolinsky H

   (1985) Spin-glass models of neural networks

   Physical Review. A, General Physics 32:1007–1018.

   https://doi.org/10.1103/physreva.32.1007

   • PubMed
   • Google Scholar
6. 1. Amit DJ
   2. Brunel N

   (1997) Model of global spontaneous activity and local structured activity during delay periods in the cerebral cortex

   Cerebral Cortex (New York, N.Y) 7:237–252.

   https://doi.org/10.1093/cercor/7.3.237

   • PubMed
   • Google Scholar
7. 1. Anderson DJ
   2. Perona P

   (2014) Toward a science of computational ethology

   Neuron 84:18–31.

   https://doi.org/10.1016/j.neuron.2014.09.005

   • PubMed
   • Google Scholar
8. 1. Aon MA
   2. Roussel MR
   3. Cortassa S
   4. O’Rourke B
   5. Murray DB
   6. Beckmann M
   7. Lloyd D

   (2008) The scale-free dynamics of eukaryotic cells

   PLOS ONE 3:e3624.

   https://doi.org/10.1371/journal.pone.0003624

   • PubMed
   • Google Scholar
9. Preprint

   1. Ayroles JF
   2. Buchanan SM
   3. Jenney C
   4. Skutt-Kakaria K
   5. Grenier J
   6. Clark AG
   7. Hartl DL
   8. de Bivort BL

   (2014) Behavioral Individuality Reveals Genetic Control of Phenotypic Variability

   bioRxiv.

   https://doi.org/10.1101/009027

   • Google Scholar
10. 1. Barthas F
    2. Kwan AC

    (2017) Secondary Motor Cortex: Where ‘Sensory’ Meets ‘Motor’ in the Rodent Frontal Cortex

    Trends in Neurosciences 40:181–193.

    https://doi.org/10.1016/j.tins.2016.11.006

    • PubMed
    • Google Scholar
11. Book

    1. Bateson PPG
    2. Hinde RA

    (1976)

    Growing Points Ethology

    Cambridge University Press.

    • Google Scholar
12. 1. Batty E
    2. Whiteway M
    3. Saxena S
    4. Biderman D
    5. Abe T
    6. Musall S
    7. Gillis W
    8. Markowitz J
    9. Churchland A
    10. Cunningham JP

    (2019)

    BehaveNet: nonlinear embedding and Bayesian neural decoding of behavioral videos

    NeurIPS.

    • Google Scholar
13. 1. Ben Arous J
    2. Laffont S
    3. Chatenay D

    (2009) Molecular and sensory basis of a food related two-state behavior in C. elegans

    PLOS ONE 4:e7584.

    https://doi.org/10.1371/journal.pone.0007584

    • PubMed
    • Google Scholar
14. 1. Berman GJ
    2. Choi DM
    3. Bialek W
    4. Shaevitz JW

    (2014) Mapping the stereotyped behaviour of freely moving fruit flies

    Journal of The Royal Society Interface 11:20140672.

    https://doi.org/10.1098/rsif.2014.0672

    • PubMed
    • Google Scholar
15. 1. Berman GJ
    2. Bialek W
    3. Shaevitz JW

    (2016) Predictability and hierarchy in Drosophila behavior

    PNAS 113:11943–11948.

    https://doi.org/10.1073/pnas.1607601113

    • Google Scholar
16. 1. Berman GJ

    (2018) Measuring behavior across scales

    BMC Biology 16:1–11.

    https://doi.org/10.1186/s12915-018-0494-7

    • PubMed
    • Google Scholar
17. 1. Bernacchia A
    2. Seo H
    3. Lee D
    4. Wang XJ

    (2011) A reservoir of time constants for memory traces in cortical neurons

    Nature Neuroscience 14:366–372.

    https://doi.org/10.1038/nn.2752

    • PubMed
    • Google Scholar
18. 1. Berridge KC
    2. Fentress JC
    3. Parr H

    (1987) Natural syntax rules control action sequence of rats

    Behavioural Brain Research 23:59–68.

    https://doi.org/10.1016/0166-4328(87)90242-7

    • PubMed
    • Google Scholar
19. 1. Bouchaud JP

    (1992) Weak ergodicity breaking and aging in disordered systems

    Journal de Physique I 2:1705–1713.

    https://doi.org/10.1051/jp1:1992238

    • Google Scholar
20. Book

    1. Bowman GR
    2. Pande VS
    3. Noé F

    (2013) An Introduction to Markov State Models and Their Application to Long Timescale Molecular Simulation, Advances in Experimental Medicine and Biology

    Springer Science & Business Media.

    https://doi.org/10.1007/978-94-007-7606-7

    • Google Scholar
21. 1. Brinkman BAW
    2. Yan H
    3. Maffei A
    4. Park IM
    5. Fontanini A
    6. Wang J
    7. La Camera G

    (2022) Metastable dynamics of neural circuits and networks

    Applied Physics Reviews 9:011313.

    https://doi.org/10.1063/5.0062603

    • PubMed
    • Google Scholar
22. 1. Brown AEX
    2. Yemini EI
    3. Grundy LJ
    4. Jucikas T
    5. Schafer WR

    (2013) A dictionary of behavioral motifs reveals clusters of genes affecting Caenorhabditis elegans locomotion

    PNAS 110:791–796.

    https://doi.org/10.1073/pnas.1211447110

    • PubMed
    • Google Scholar
23. 1. Brown AEX
    2. de Bivort B

    (2018) Ethology as a physical science

    Nature Physics 14:653–657.

    https://doi.org/10.1038/s41567-018-0093-0

    • Google Scholar
24. 1. Brunel N

    (2000) Dynamics of sparsely connected networks of excitatory and inhibitory spiking neurons

    Journal of Computational Neuroscience 8:183–208.

    https://doi.org/10.1023/a:1008925309027

    • PubMed
    • Google Scholar
25. 1. Buchanan SM
    2. Kain JS
    3. de Bivort BL

    (2015) Neuronal control of locomotor handedness in Drosophila

    PNAS 112:6700–6705.

    https://doi.org/10.1073/pnas.1500804112

    • PubMed
    • Google Scholar
26. 1. Bucher D
    2. Prinz AA
    3. Marder E

    (2005) Animal-to-Animal Variability in Motor Pattern Production in Adults and during Growth

    Journal of Neuroscience 25:1611–1619.

    https://doi.org/10.1523/JNEUROSCI.3679-04.2005

    • PubMed
    • Google Scholar
27. 1. Calhoun AJ
    2. Pillow JW
    3. Murthy M

    (2019) Unsupervised identification of the internal states that shape natural behavior

    Nature Neuroscience 22:2040–2049.

    https://doi.org/10.1038/s41593-019-0533-x

    • PubMed
    • Google Scholar
28. 1. Cande J
    2. Namiki S
    3. Qiu J
    4. Korff W
    5. Card GM
    6. Shaevitz JW
    7. Stern DL
    8. Berman GJ

    (2018) Optogenetic dissection of descending behavioral control in Drosophila

    eLife 7:e34275.

    https://doi.org/10.7554/eLife.34275

    • PubMed
    • Google Scholar
29. 1. Cao R
    2. Pastukhov A
    3. Mattia M
    4. Braun J

    (2016) Collective Activity of Many Bistable Assemblies Reproduces Characteristic Dynamics of Multistable Perception

    Journal of Neuroscience 36:6957–6972.

    https://doi.org/10.1523/JNEUROSCI.4626-15.2016

    • PubMed
    • Google Scholar
30. 1. Cavanagh SE
    2. Wallis JD
    3. Kennerley SW
    4. Hunt LT

    (2016) Autocorrelation structure at rest predicts value correlates of single neurons during reward-guided choice

    eLife 5:e18937.

    https://doi.org/10.7554/eLife.18937

    • PubMed
    • Google Scholar
31. Preprint

    1. Cazettes F
    2. Murakami M
    3. Renart A
    4. Mainen Z

    (2021) Reservoir of Decision Strategies in the Mouse Brain

    bioRxiv.

    https://doi.org/10.1101/2021.04.01.438090

    • Google Scholar
32. 1. Chance FS
    2. Abbott LF
    3. Reyes AD

    (2002) Gain modulation from background synaptic input

    Neuron 35:773–782.

    https://doi.org/10.1016/s0896-6273(02)00820-6

    • PubMed
    • Google Scholar
33. 1. Chaudhuri R
    2. Bernacchia A
    3. Wang XJ

    (2014) A diversity of localized timescales in network activity

    eLife 3:e01239.

    https://doi.org/10.7554/eLife.01239

    • PubMed
    • Google Scholar
34. 1. Churchland MM
    2. Afshar A
    3. Shenoy KV

    (2006) A central source of movement variability

    Neuron 52:1085–1096.

    https://doi.org/10.1016/j.neuron.2006.10.034

    • PubMed
    • Google Scholar
35. 1. Churgin MA
    2. Jung SK
    3. Yu CC
    4. Chen X
    5. Raizen DM
    6. Fang-Yen C

    (2017) Longitudinal imaging of Caenorhabditis elegans in a microfabricated device reveals variation in behavioral decline during aging

    eLife 6:e26652.

    https://doi.org/10.7554/eLife.26652

    • PubMed
    • Google Scholar
36. 1. Coen P
    2. Clemens J
    3. Weinstein AJ
    4. Pacheco DA
    5. Deng Y
    6. Murthy M

    (2014) Dynamic sensory cues shape song structure in Drosophila

    Nature 507:233–237.

    https://doi.org/10.1038/nature13131

    • PubMed
    • Google Scholar
37. 1. Cohen Y
    2. Shen J
    3. Semu D
    4. Leman DP
    5. Liberti WA
    6. Perkins LN
    7. Liberti DC
    8. Kotton DN
    9. Gardner TJ

    (2020) Hidden neural states underlie canary song syntax

    Nature 582:539–544.

    https://doi.org/10.1038/s41586-020-2397-3

    • PubMed
    • Google Scholar
38. 1. Corrales-Carvajal VM
    2. Faisal AA
    3. Ribeiro C

    (2016) Internal states drive nutrient homeostasis by modulating exploration-exploitation trade-off

    eLife 5:e19920.

    https://doi.org/10.7554/eLife.19920

    • PubMed
    • Google Scholar
39. 1. Costa AC
    2. Ahamed T
    3. Stephens GJ

    (2019) Adaptive, locally linear models of complex dynamics

    PNAS 116:1501–1510.

    https://doi.org/10.1073/pnas.1813476116

    • PubMed
    • Google Scholar
40. Preprint

    1. Costa AC
    2. Ahamed T
    3. Jordan D
    4. Stephens GJ

    (2021) Maximally Predictive Ensemble Dynamics from Data

    Biophysics.

    https://doi.org/10.1101/2021.05.26.445816

    • Google Scholar
41. 1. Darshan R
    2. Wood WE
    3. Peters S
    4. Leblois A
    5. Hansel D

    (2017) A canonical neural mechanism for behavioral variability

    Nature Communications 8:1–13.

    https://doi.org/10.1038/ncomms15415

    • PubMed
    • Google Scholar
42. 1. Datta SR
    2. Anderson DJ
    3. Branson K
    4. Perona P
    5. Leifer A

    (2019) Computational Neuroethology: A Call to Action

    Neuron 104:11–24.

    https://doi.org/10.1016/j.neuron.2019.09.038

    • PubMed
    • Google Scholar
43. 1. Dawkins R

    (1976)

    Hierarchical organisation: A candidate principle for ethology

    Growing Points in Ethology 7:54.

    • Google Scholar
44. 1. Deco G
    2. Hugues E

    (2012) Neural network mechanisms underlying stimulus driven variability reduction

    PLOS Computational Biology 8:e1002395.

    https://doi.org/10.1371/journal.pcbi.1002395

    • PubMed
    • Google Scholar
45. 1. Deutsch D
    2. Pacheco D
    3. Encarnacion-Rivera L
    4. Pereira T
    5. Fathy R
    6. Clemens J
    7. Girardin C
    8. Calhoun A
    9. Ireland E
    10. Burke A
    11. Dorkenwald S
    12. McKellar C
    13. Macrina T
    14. Lu R
    15. Lee K
    16. Kemnitz N
    17. Ih D
    18. Castro M
    19. Halageri A
    20. Jordan C
    21. Silversmith W
    22. Wu J
    23. Seung HS
    24. Murthy M

    (2020) The neural basis for a persistent internal state in Drosophila females

    eLife 9:e59502.

    https://doi.org/10.7554/eLife.59502

    • PubMed
    • Google Scholar
46. 1. Dhawale AK
    2. Smith MA
    3. Ölveczky BP

    (2017) The Role of Variability in Motor Learning

    Annual Review of Neuroscience 40:479–498.

    https://doi.org/10.1146/annurev-neuro-072116-031548

    • PubMed
    • Google Scholar
47. 1. Dissel S
    2. Hansen CN
    3. Özkaya Ö
    4. Hemsley M
    5. Kyriacou CP
    6. Rosato E

    (2014) The logic of circadian organization in Drosophila

    Current Biology 24:2257–2266.

    https://doi.org/10.1016/j.cub.2014.08.023

    • PubMed
    • Google Scholar
48. Conference

    1. Duncker L
    2. Bohner G
    3. Boussard J
    4. Sahani M

    (2019)

    Learning interpretable continuous-time models of latent stochastic dynamical systems

    International Conference on Machine Learning PMLR. pp. 1726–1734.

    • Google Scholar
49. 1. Eliades SJ
    2. Miller CT

    (2017) Marmoset vocal communication: Behavior and neurobiology

    Developmental Neurobiology 77:286–299.

    https://doi.org/10.1002/dneu.22464

    • PubMed
    • Google Scholar
50. 1. Emmerling M
    2. Dauner M
    3. Ponti A
    4. Fiaux J
    5. Hochuli M
    6. Szyperski T
    7. Wüthrich K
    8. Bailey JE
    9. Sauer U

    (2002) Metabolic Flux Responses to Pyruvate Kinase Knockout in Escherichia coli

    Journal of Bacteriology 184:152–164.

    https://doi.org/10.1128/JB.184.1.152-164.2002

    • PubMed
    • Google Scholar
51. 1. Fan D
    2. Rossi MA
    3. Yin HH

    (2012) Mechanisms of Action Selection and Timing in Substantia Nigra Neurons

    Journal of Neuroscience 32:5534–5548.

    https://doi.org/10.1523/JNEUROSCI.5924-11.2012

    • PubMed
    • Google Scholar
52. 1. Findley TM
    2. Wyrick DG
    3. Cramer JL
    4. Brown MA
    5. Holcomb B
    6. Attey R
    7. Yeh D
    8. Monasevitch E
    9. Nouboussi N
    10. Cullen I
    11. Songco JO
    12. King JF
    13. Ahmadian Y
    14. Smear MC

    (2021) Sniff-synchronized, gradient-guided olfactory search by freely moving mice

    eLife 10:e58523.

    https://doi.org/10.7554/eLife.58523

    • PubMed
    • Google Scholar
53. 1. Finkelstein A
    2. Fontolan L
    3. Economo MN
    4. Li N
    5. Romani S
    6. Svoboda K

    (2021) Attractor dynamics gate cortical information flow during decision-making

    Nature Neuroscience 24:843–850.

    https://doi.org/10.1038/s41593-021-00840-6

    • PubMed
    • Google Scholar
54. 1. Flavell SW
    2. Pokala N
    3. Macosko EZ
    4. Albrecht DR
    5. Larsch J
    6. Bargmann CI

    (2013) Serotonin and the neuropeptide PDF initiate and extend opposing behavioral states in C. elegans

    Cell 154:1023–1035.

    https://doi.org/10.1016/j.cell.2013.08.001

    • PubMed
    • Google Scholar
55. 1. Fonseca MS
    2. Murakami M
    3. Mainen ZF

    (2015) Activation of dorsal raphe serotonergic neurons promotes waiting but is not reinforcing

    Current Biology 25:306–315.

    https://doi.org/10.1016/j.cub.2014.12.002

    • PubMed
    • Google Scholar
56. 1. Fourcaud N
    2. Brunel N

    (2002) Dynamics of the firing probability of noisy integrate-and-fire neurons

    Neural Computation 14:2057–2110.

    https://doi.org/10.1162/089976602320264015

    • PubMed
    • Google Scholar
57. 1. Funahashi S
    2. Bruce CJ
    3. Goldman-Rakic PS

    (1989) Mnemonic coding of visual space in the monkey’s dorsolateral prefrontal cortex

    Journal of Neurophysiology 61:331–349.

    https://doi.org/10.1152/jn.1989.61.2.331

    • PubMed
    • Google Scholar
58. 1. Fuster JM
    2. Jervey JP

    (1981) Inferotemporal neurons distinguish and retain behaviorally relevant features of visual stimuli

    Science (New York, N.Y.) 212:952–955.

    https://doi.org/10.1126/science.7233192

    • PubMed
    • Google Scholar
59. 1. Geddes CE
    2. Li H
    3. Jin X

    (2018) Optogenetic Editing Reveals the Hierarchical Organization of Learned Action Sequences

    Cell 174:32–43.

    https://doi.org/10.1016/j.cell.2018.06.012

    • PubMed
    • Google Scholar
60. 1. Genkin M
    2. Engel TA

    (2020) Moving beyond generalization to accurate interpretation of flexible models

    Nature Machine Intelligence 2:674–683.

    https://doi.org/10.1038/s42256-020-00242-6

    • Google Scholar
61. 1. Gjorgjieva J
    2. Drion G
    3. Marder E

    (2016) Computational implications of biophysical diversity and multiple timescales in neurons and synapses for circuit performance

    Current Opinion in Neurobiology 37:44–52.

    https://doi.org/10.1016/j.conb.2015.12.008

    • PubMed
    • Google Scholar
62. 1. Gomez-Marin A
    2. Stephens GJ
    3. Brown AEX

    (2016) Hierarchical compression of Caenorhabditis elegans locomotion reveals phenotypic differences in the organization of behaviour

    Journal of the Royal Society, Interface 13:20160466.

    https://doi.org/10.1098/rsif.2016.0466

    • PubMed
    • Google Scholar
63. 1. Guo ZV
    2. Inagaki HK
    3. Daie K
    4. Druckmann S
    5. Gerfen CR
    6. Svoboda K

    (2017) Maintenance of persistent activity in a frontal thalamocortical loop

    Nature 545:181–186.

    https://doi.org/10.1038/nature22324

    • PubMed
    • Google Scholar
64. 1. Guo K
    2. Yamawaki N
    3. Svoboda K
    4. Shepherd GMG

    (2018) Anterolateral Motor Cortex Connects with a Medial Subdivision of Ventromedial Thalamus through Cell Type-Specific Circuits, Forming an Excitatory Thalamo-Cortico-Thalamic Loop via Layer 1 Apical Tuft Dendrites of Layer 5B Pyramidal Tract Type Neurons

    The Journal of Neuroscience 38:8787–8797.

    https://doi.org/10.1523/JNEUROSCI.1333-18.2018

    • PubMed
    • Google Scholar
65. Preprint

    1. Gupta S
    2. Gomez-Marin A

    (2019) A Context-Free Grammar for Caenorhabditis Elegans Behavior

    bioRxiv.

    https://doi.org/10.1101/708891

    • Google Scholar
66. 1. Hallinen KM
    2. Dempsey R
    3. Scholz M
    4. Yu X
    5. Linder A
    6. Randi F
    7. Sharma AK
    8. Shaevitz JW
    9. Leifer AM

    (2021) Decoding locomotion from population neural activity in moving C. elegans

    eLife 10:e66135.

    https://doi.org/10.7554/eLife.66135

    • PubMed
    • Google Scholar
67. 1. Hänggi P
    2. Talkner P
    3. Borkovec M

    (1990) Reaction-rate theory: fifty years after Kramers

    Reviews of Modern Physics 62:251–341.

    https://doi.org/10.1103/RevModPhys.62.251

    • Google Scholar
68. 1. Hastings MH
    2. Maywood ES
    3. Brancaccio M

    (2018) Generation of circadian rhythms in the suprachiasmatic nucleus

    Nature Reviews. Neuroscience 19:453–469.

    https://doi.org/10.1038/s41583-018-0026-z

    • PubMed
    • Google Scholar
69. 1. Honegger K
    2. de Bivort B

    (2018) Stochasticity, individuality and behavior

    Current Biology 28:R8–R12.

    https://doi.org/10.1016/j.cub.2017.11.058

    • PubMed
    • Google Scholar
70. 1. Hong W
    2. Kennedy A
    3. Burgos-Artizzu XP
    4. Zelikowsky M
    5. Navonne SG
    6. Perona P
    7. Anderson DJ

    (2015) Automated measurement of mouse social behaviors using depth sensing, video tracking, and machine learning

    PNAS 112:E5351–E5360.

    https://doi.org/10.1073/pnas.1515982112

    • PubMed
    • Google Scholar
71. 1. Hopfield JJ

    (1982) Neural networks and physical systems with emergent collective computational abilities

    PNAS 79:2554–2558.

    https://doi.org/10.1073/pnas.79.8.2554

    • Google Scholar
72. 1. Hopfield JJ

    (1984) Neurons with graded response have collective computational properties like those of two-state neurons

    PNAS 81:3088–3092.

    https://doi.org/10.1073/pnas.81.10.3088

    • Google Scholar
73. 1. Hoy JL
    2. Yavorska I
    3. Wehr M
    4. Niell CM

    (2016) Vision Drives Accurate Approach Behavior during Prey Capture in Laboratory Mice

    Current Biology 26:3046–3052.

    https://doi.org/10.1016/j.cub.2016.09.009

    • PubMed
    • Google Scholar
74. 1. Huang C
    2. Doiron B

    (2017) Once upon a (slow) time in the land of recurrent neuronal networks

    Current Opinion in Neurobiology 46:31–38.

    https://doi.org/10.1016/j.conb.2017.07.003

    • PubMed
    • Google Scholar
75. 1. Inagaki HK
    2. Fontolan L
    3. Romani S
    4. Svoboda K

    (2019) Discrete attractor dynamics underlies persistent activity in the frontal cortex

    Nature 566:212–217.

    https://doi.org/10.1038/s41586-019-0919-7

    • PubMed
    • Google Scholar
76. 1. Inagaki HK
    2. Chen S
    3. Ridder MC
    4. Sah P
    5. Li N
    6. Yang Z
    7. Hasanbegovic H
    8. Gao Z
    9. Gerfen CR
    10. Svoboda K

    (2022) A midbrain-thalamus-cortex circuit reorganizes cortical dynamics to initiate movement

    Cell 185:1065–1081.

    https://doi.org/10.1016/j.cell.2022.02.006

    • PubMed
    • Google Scholar
77. 1. Jaramillo S
    2. Zador AM

    (2011) The auditory cortex mediates the perceptual effects of acoustic temporal expectation

    Nature Neuroscience 14:246–251.

    https://doi.org/10.1038/nn.2688

    • PubMed
    • Google Scholar
78. 1. Johnson RE
    2. Linderman S
    3. Panier T
    4. Wee CL
    5. Song E
    6. Herrera KJ
    7. Miller A
    8. Engert F

    (2020) Probabilistic Models of Larval Zebrafish Behavior Reveal Structure on Many Scales

    Current Biology 30:70–82.

    https://doi.org/10.1016/j.cub.2019.11.026

    • PubMed
    • Google Scholar
79. 1. Jones LM
    2. Fontanini A
    3. Sadacca BF
    4. Miller P
    5. Katz DB

    (2007) Natural stimuli evoke dynamic sequences of states in sensory cortical ensembles

    PNAS 104:18772–18777.

    https://doi.org/10.1073/pnas.0705546104

    • PubMed
    • Google Scholar
80. 1. Joshua M
    2. Medina JF
    3. Lisberger SG

    (2013) Diversity of neural responses in the brainstem during smooth pursuit eye movements constrains the circuit mechanisms of neural integration

    The Journal of Neuroscience 33:6633–6647.

    https://doi.org/10.1523/JNEUROSCI.3732-12.2013

    • PubMed
    • Google Scholar
81. 1. Juavinett AL
    2. Bekheet G
    3. Churchland AK

    (2019) Chronically implanted Neuropixels probes enable high-yield recordings in freely moving mice

    eLife 8:e47188.

    https://doi.org/10.7554/eLife.47188

    • PubMed
    • Google Scholar
82. 1. Kao MH
    2. Doupe AJ
    3. Brainard MS

    (2005) Contributions of an avian basal ganglia-forebrain circuit to real-time modulation of song

    Nature 433:638–643.

    https://doi.org/10.1038/nature03127

    • PubMed
    • Google Scholar
83. 1. Kaplan HS
    2. Salazar Thula O
    3. Khoss N
    4. Zimmer M

    (2020) Nested Neuronal Dynamics Orchestrate a Behavioral Hierarchy across Timescales

    Neuron 105:562–576.

    https://doi.org/10.1016/j.neuron.2019.10.037

    • PubMed
    • Google Scholar
84. 1. Kawai R
    2. Markman T
    3. Poddar R
    4. Ko R
    5. Fantana AL
    6. Dhawale AK
    7. Kampff AR
    8. Ölveczky BP

    (2015) Motor cortex is required for learning but not for executing a motor skill

    Neuron 86:800–812.

    https://doi.org/10.1016/j.neuron.2015.03.024

    • PubMed
    • Google Scholar
85. 1. Kiani R
    2. Cueva CJ
    3. Reppas JB
    4. Peixoto D
    5. Ryu SI
    6. Newsome WT

    (2015) Natural Grouping of Neural Responses Reveals Spatially Segregated Clusters in Prearcuate Cortex

    Neuron 85:1359–1373.

    https://doi.org/10.1016/j.neuron.2015.02.014

    • PubMed
    • Google Scholar
86. 1. Killeen PR
    2. Fetterman JG

    (1988) A behavioral theory of timing

    Psychological Review 95:274–295.

    https://doi.org/10.1037/0033-295x.95.2.274

    • PubMed
    • Google Scholar
87. Preprint

    1. Klibaite U
    2. Kislin M
    3. Verpeut JL
    4. Sun X
    5. Shaevitz JW
    6. Wang SSH

    (2021) Deep Phenotyping Reveals Movement Phenotypes in Mouse Neurodevelopmental Models

    In Review.

    https://doi.org/10.21203/rs.3.rs-798847/v1

    • Google Scholar
88. 1. Koch KW
    2. Fuster JM

    (1989) Unit activity in monkey parietal cortex related to haptic perception and temporary memory

    Experimental Brain Research 76:292–306.

    https://doi.org/10.1007/BF00247889

    • PubMed
    • Google Scholar
89. 1. La Camera G
    2. Fontanini A
    3. Mazzucato L

    (2019) Cortical computations via metastable activity

    Current Opinion in Neurobiology 58:37–45.

    https://doi.org/10.1016/j.conb.2019.06.007

    • PubMed
    • Google Scholar
90. Preprint

    1. Lauer J
    2. Zhou M
    3. Ye S
    4. Menegas W
    5. Nath T
    6. Rahman MM
    7. Di Santo V
    8. Soberanes D
    9. Feng G
    10. Murthy VN
    11. Lauder G
    12. Dulac C
    13. Mathis MW
    14. Mathis A

    (2021) Multi-Animal Pose Estimation and Tracking with DeepLabCut

    Animal Behavior and Cognition.

    https://doi.org/10.1101/2021.04.30.442096

    • Google Scholar
91. 1. Lee W-CA
    2. Bonin V
    3. Reed M
    4. Graham BJ
    5. Hood G
    6. Glattfelder K
    7. Reid RC

    (2016) Anatomy and function of an excitatory network in the visual cortex

    Nature 532:370–374.

    https://doi.org/10.1038/nature17192

    • Google Scholar
92. 1. Li N
    2. Daie K
    3. Svoboda K
    4. Druckmann S

    (2016) Robust neuronal dynamics in premotor cortex during motor planning

    Nature 532:459–464.

    https://doi.org/10.1038/nature17643

    • PubMed
    • Google Scholar
93. 1. Lim S
    2. McKee JL
    3. Woloszyn L
    4. Amit Y
    5. Freedman DJ
    6. Sheinberg DL
    7. Brunel N

    (2015) Inferring learning rules from distributions of firing rates in cortical neurons

    Nature Neuroscience 18:1804–1810.

    https://doi.org/10.1038/nn.4158

    • PubMed
    • Google Scholar
94. Book

    1. Linderman S
    2. Johnson M
    3. Miller A
    4. Adams R
    5. Blei D
    6. Paninski L

    (2017) Bayesian learning and inference in recurrent switching linear dynamical systems

    In: Singh A, Zhu J, editors. Proceedings of the 20th International Conference on Artificial Intelligence and Statistics. PMLR. pp. 914–922.

    https://doi.org/10.1109/CAMSAP.2017.8313132

    • Google Scholar
95. 1. Linderman S
    2. Nichols A
    3. Blei D
    4. Zimmer M
    5. Paninski L

    (2019) Hierarchical Recurrent State Space Models Reveal Discrete and Continuous Dynamics of Neural Activity in C. Elegans

    Neuroscience 621540.

    https://doi.org/10.1101/621540

    • Google Scholar
96. 1. Litwin-Kumar A
    2. Doiron B

    (2012) Slow dynamics and high variability in balanced cortical networks with clustered connections

    Nature Neuroscience 15:1498–1505.

    https://doi.org/10.1038/nn.3220

    • PubMed
    • Google Scholar
97. 1. Litwin-Kumar A
    2. Doiron B

    (2014) Formation and maintenance of neuronal assemblies through synaptic plasticity

    Nature Communications 5:1–12.

    https://doi.org/10.1038/ncomms6319

    • PubMed
    • Google Scholar
98. Preprint

    1. Logiaco L
    2. Escola GS

    (2020) Thalamocortical Motor Circuit Insights for More Robust Hierarchical Control of Complex Sequences

    ArXiv Preprint.

    https://arxiv.org/abs/2006.13332

    • Google Scholar
99. 1. Logiaco L
    2. Abbott LF
    3. Escola S

    (2021) Thalamic control of cortical dynamics in a model of flexible motor sequencing

    Cell Reports 35:109090.

    https://doi.org/10.1016/j.celrep.2021.109090

    • PubMed
    • Google Scholar
100. 1. Lottem E
     2. Banerjee D
     3. Vertechi P
     4. Sarra D
     5. Lohuis MO
     6. Mainen ZF

     (2018) Activation of serotonin neurons promotes active persistence in a probabilistic foraging task

     Nature Communications 9:1–12.

     https://doi.org/10.1038/s41467-018-03438-y

     • PubMed
     • Google Scholar
101. 1. Loveless J
     2. Lagogiannis K
     3. Webb B

     (2019) Modelling the mechanics of exploration in larval Drosophila

     PLOS Computational Biology 15:e1006635.

     https://doi.org/10.1371/journal.pcbi.1006635

     • PubMed
     • Google Scholar
102. 1. Maboudi K
     2. Ackermann E
     3. de Jong LW
     4. Pfeiffer BE
     5. Foster D
     6. Diba K
     7. Kemere C

     (2018) Uncovering temporal structure in hippocampal output patterns

     eLife 7:e34467.

     https://doi.org/10.7554/eLife.34467

     • PubMed
     • Google Scholar
103. 1. MacDougall M
     2. Nummela SU
     3. Coop S
     4. Disney A
     5. Mitchell JF
     6. Miller CT

     (2016) Optogenetic manipulation of neural circuits in awake marmosets

     Journal of Neurophysiology 116:1286–1294.

     https://doi.org/10.1152/jn.00197.2016

     • PubMed
     • Google Scholar
104. 1. Magnasco MO
     2. Piro O
     3. Cecchi GA

     (2009) Self-tuned critical anti-Hebbian networks

     Physical Review Letters 102:258102.

     https://doi.org/10.1103/PhysRevLett.102.258102

     • PubMed
     • Google Scholar
105. 1. Markowitz JE
     2. Ivie E
     3. Kligler L
     4. Gardner TJ

     (2013) Long-range order in canary song

     PLOS Computational Biology 9:e1003052.

     https://doi.org/10.1371/journal.pcbi.1003052

     • PubMed
     • Google Scholar
106. 1. Markowitz JE
     2. Gillis WF
     3. Beron CC
     4. Neufeld SQ
     5. Robertson K
     6. Bhagat ND
     7. Peterson RE
     8. Peterson E
     9. Hyun M
     10. Linderman SW
     11. Sabatini BL
     12. Datta SR

     (2018) The Striatum Organizes 3D Behavior via Moment-to-Moment Action Selection

     Cell 174:44–58.

     https://doi.org/10.1016/j.cell.2018.04.019

     • PubMed
     • Google Scholar
107. 1. Marques JC
     2. Lackner S
     3. Félix R
     4. Orger MB

     (2018) Structure of the Zebrafish Locomotor Repertoire Revealed with Unsupervised Behavioral Clustering

     Current Biology 28:181–195.

     https://doi.org/10.1016/j.cub.2017.12.002

     • PubMed
     • Google Scholar
108. 1. Marshall JD
     2. Aldarondo DE
     3. Dunn TW
     4. Wang WL
     5. Berman GJ
     6. Ölveczky BP

     (2021) Continuous Whole-Body 3D Kinematic Recordings across the Rodent Behavioral Repertoire

     Neuron 109:420–437.

     https://doi.org/10.1016/j.neuron.2020.11.016

     • PubMed
     • Google Scholar
109. 1. Marshel JH
     2. Kim YS
     3. Machado TA
     4. Quirin S
     5. Benson B
     6. Kadmon J
     7. Raja C
     8. Chibukhchyan A
     9. Ramakrishnan C
     10. Inoue M
     11. Shane JC
     12. McKnight DJ
     13. Yoshizawa S
     14. Kato HE
     15. Ganguli S
     16. Deisseroth K

     (2019) Cortical layer-specific critical dynamics triggering perception

     Science (New York, N.Y.) 365:eaaw5202.

     https://doi.org/10.1126/science.aaw5202

     • PubMed
     • Google Scholar
110. 1. Martí D
     2. Brunel N
     3. Ostojic S

     (2018) Correlations between synapses in pairs of neurons slow down dynamics in randomly connected neural networks

     Physical Review. E 97:062314.

     https://doi.org/10.1103/PhysRevE.97.062314

     • PubMed
     • Google Scholar
111. 1. Mascaro M
     2. Amit DJ

     (1999)

     Effective neural response function for collective population states

     Network (Bristol, England) 10:351–373.

     • PubMed
     • Google Scholar
112. 1. Mathis A
     2. Mamidanna P
     3. Cury KM
     4. Abe T
     5. Murthy VN
     6. Mathis MW
     7. Bethge M

     (2018) DeepLabCut: markerless pose estimation of user-defined body parts with deep learning

     Nature Neuroscience 21:1281–1289.

     https://doi.org/10.1038/s41593-018-0209-y

     • PubMed
     • Google Scholar
113. 1. Mattia M
     2. Sanchez-Vives MV

     (2012) Exploring the spectrum of dynamical regimes and timescales in spontaneous cortical activity

     Cognitive Neurodynamics 6:239–250.

     https://doi.org/10.1007/s11571-011-9179-4

     • PubMed
     • Google Scholar
114. 1. Mattia M
     2. Pani P
     3. Mirabella G
     4. Costa S
     5. Del Giudice P
     6. Ferraina S

     (2013) Heterogeneous attractor cell assemblies for motor planning in premotor cortex

     The Journal of Neuroscience 33:11155–11168.

     https://doi.org/10.1523/JNEUROSCI.4664-12.2013

     • PubMed
     • Google Scholar
115. 1. Mazzucato L
     2. Fontanini A
     3. La Camera G

     (2015) Dynamics of multistable states during ongoing and evoked cortical activity

     The Journal of Neuroscience 35:8214–8231.

     https://doi.org/10.1523/JNEUROSCI.4819-14.2015

     • PubMed
     • Google Scholar
116. 1. Mazzucato L
     2. La Camera G
     3. Fontanini A

     (2019) Expectation-induced modulation of metastable activity underlies faster coding of sensory stimuli

     Nature Neuroscience 22:787–796.

     https://doi.org/10.1038/s41593-019-0364-9

     • PubMed
     • Google Scholar
117. 1. McKenzie S
     2. Frank AJ
     3. Kinsky NR
     4. Porter B
     5. Rivière PD
     6. Eichenbaum H

     (2014) Hippocampal representation of related and opposing memories develop within distinct, hierarchically organized neural schemas

     Neuron 83:202–215.

     https://doi.org/10.1016/j.neuron.2014.05.019

     • PubMed
     • Google Scholar
118. 1. Michaiel AM
     2. Abe ET
     3. Niell CM

     (2020) Dynamics of gaze control during prey capture in freely moving mice

     eLife 9:e57458.

     https://doi.org/10.7554/eLife.57458

     • PubMed
     • Google Scholar
119. 1. Miller P
     2. Katz DB

     (2010) Stochastic transitions between neural states in taste processing and decision-making

     The Journal of Neuroscience 30:2559–2570.

     https://doi.org/10.1523/JNEUROSCI.3047-09.2010

     • PubMed
     • Google Scholar
120. 1. Miller JK
     2. Ayzenshtat I
     3. Carrillo-Reid L
     4. Yuste R

     (2014) Visual stimuli recruit intrinsically generated cortical ensembles

     PNAS 111:E4053–E4061.

     https://doi.org/10.1073/pnas.1406077111

     • PubMed
     • Google Scholar
121. 1. Miri A
     2. Daie K
     3. Arrenberg AB
     4. Baier H
     5. Aksay E
     6. Tank DW

     (2011) Spatial gradients and multidimensional dynamics in a neural integrator circuit

     Nature Neuroscience 14:1150–1159.

     https://doi.org/10.1038/nn.2888

     • PubMed
     • Google Scholar
122. 1. Mischel W
     2. Ebbesen EB

     (1970) Attention in delay of gratification

     Journal of Personality and Social Psychology 16:329–337.

     https://doi.org/10.1037/h0029815

     • Google Scholar
123. 1. Miyashita Y
     2. Chang HS

     (1988) Neuronal correlate of pictorial short-term memory in the primate temporal cortex

     Nature 331:68–70.

     https://doi.org/10.1038/331068a0

     • PubMed
     • Google Scholar
124. 1. Moreno-Bote R
     2. Beck J
     3. Kanitscheider I
     4. Pitkow X
     5. Latham P
     6. Pouget A

     (2014) Information-limiting correlations

     Nature Neuroscience 17:1410–1417.

     https://doi.org/10.1038/nn.3807

     • PubMed
     • Google Scholar
125. 1. Murakami M
     2. Vicente MI
     3. Costa GM
     4. Mainen ZF

     (2014) Neural antecedents of self-initiated actions in secondary motor cortex

     Nature Neuroscience 17:1574–1582.

     https://doi.org/10.1038/nn.3826

     • PubMed
     • Google Scholar
126. 1. Murakami M
     2. Shteingart H
     3. Loewenstein Y
     4. Mainen ZF

     (2017) Distinct Sources of Deterministic and Stochastic Components of Action Timing Decisions in Rodent Frontal Cortex

     Neuron 94:908–919.

     https://doi.org/10.1016/j.neuron.2017.04.040

     • PubMed
     • Google Scholar
127. 1. Murray JD
     2. Bernacchia A
     3. Freedman DJ
     4. Romo R
     5. Wallis JD
     6. Cai X
     7. Padoa-Schioppa C
     8. Pasternak T
     9. Seo H
     10. Lee D
     11. Wang X-J

     (2014) A hierarchy of intrinsic timescales across primate cortex

     Nature Neuroscience 17:1661–1663.

     https://doi.org/10.1038/nn.3862

     • PubMed
     • Google Scholar
128. 1. Murray JM
     2. Escola GS

     (2017) Learning multiple variable-speed sequences in striatum via cortical tutoring

     eLife 6:e26084.

     https://doi.org/10.7554/eLife.26084

     • PubMed
     • Google Scholar
129. 1. Murray J.M
     2. Escola GS

     (2020) Remembrance of things practiced with fast and slow learning in cortical and subcortical pathways

     Nature Communications 11:1–12.

     https://doi.org/10.1038/s41467-020-19788-5

     • PubMed
     • Google Scholar
130. Preprint

     1. Nassar J
     2. Linderman SW
     3. Bugallo M
     4. Park IM

     (2018) Tree-Structured Recurrent Switching Linear Dynamical Systems For Multi-Scale Modeling

     arXiv.

     https://arxiv.org/abs/1811.12386

     • Google Scholar
131. Book

     1. Nelson RJ.

     (2005)

     An introduction to behavioral endocrinology

     Sinauer Associates.

     • Google Scholar
132. 1. Nourizonoz A
     2. Zimmermann R
     3. Ho CLA
     4. Pellat S
     5. Ormen Y
     6. Prévost-Solié C
     7. Reymond G
     8. Pifferi F
     9. Aujard F
     10. Herrel A
     11. Huber D

     (2020) EthoLoop: automated closed-loop neuroethology in naturalistic environments

     Nature Methods 17:1052–1059.

     https://doi.org/10.1038/s41592-020-0961-2

     • PubMed
     • Google Scholar
133. 1. Nummela SU
     2. Jovanovic V
     3. de la Mothe L
     4. Miller CT

     (2017) Social Context-Dependent Activity in Marmoset Frontal Cortex Populations during Natural Conversations

     The Journal of Neuroscience 37:7036–7047.

     https://doi.org/10.1523/JNEUROSCI.0702-17.2017

     • PubMed
     • Google Scholar
134. 1. Olveczky BP
     2. Andalman AS
     3. Fee MS
     4. Schultz W

     (2005) Vocal Experimentation in the Juvenile Songbird Requires a Basal Ganglia Circuit

     PLOS Biology 3:e153.

     https://doi.org/10.1371/journal.pbio.0030153

     • PubMed
     • Google Scholar
135. 1. Parga N
     2. Virasoro MA

     (1986) The ultrametric organization of memories in a neural network

     Journal de Physique 47:1857–1864.

     https://doi.org/10.1051/jphys:0198600470110185700

     • Google Scholar
136. Preprint

     1. Parker PRL
     2. Abe ETT
     3. Beatie NT
     4. Leonard ESP
     5. Martins DM
     6. Sharp SL
     7. Wyrick DG
     8. Mazzucato L
     9. Niell CM

     (2021) Distance Estimation from Monocular Cues in an Ethological Visuomotor Task

     bioRxiv.

     https://doi.org/10.1101/2021.09.29.462468

     • Google Scholar
137. 1. Patke A
     2. Young MW
     3. Axelrod S

     (2020) Molecular mechanisms and physiological importance of circadian rhythms

     Nature Reviews Molecular Cell Biology 21:67–84.

     https://doi.org/10.1038/s41580-019-0179-2

     • PubMed
     • Google Scholar
138. 1. Pereira TD
     2. Aldarondo DE
     3. Willmore L
     4. Kislin M
     5. Wang SSH
     6. Murthy M
     7. Shaevitz JW

     (2019) Fast animal pose estimation using deep neural networks

     Nature Methods 16:117–125.

     https://doi.org/10.1038/s41592-018-0234-5

     • PubMed
     • Google Scholar
139. 1. Pereira TD
     2. Shaevitz JW
     3. Murthy M

     (2020a) Quantifying behavior to understand the brain

     Nature Neuroscience 23:1537–1549.

     https://doi.org/10.1038/s41593-020-00734-z

     • PubMed
     • Google Scholar
140. 1. Pereira TD
     2. Tabris N
     3. Li J
     4. Ravindranath S
     5. Papadoyannis ES
     6. Wang ZY
     7. Turner DM
     8. McKenzie-Smith G
     9. Kocher SD
     10. Falkner AL
     11. Shaevitz JW
     12. Murthy M

     (2020b) SLEAP: Multi-Animal Pose Tracking

     Animal Behavior and Cognition.

     https://doi.org/10.1101/2020.08.31.276246

     • Google Scholar
141. 1. Perin R
     2. Berger TK
     3. Markram H

     (2011) A synaptic organizing principle for cortical neuronal groups

     PNAS 108:5419–5424.

     https://doi.org/10.1073/pnas.1016051108

     • PubMed
     • Google Scholar
142. 1. Prinz AA
     2. Bucher D
     3. Marder E

     (2004) Similar network activity from disparate circuit parameters

     Nature Neuroscience 7:1345–1352.

     https://doi.org/10.1038/nn1352

     • PubMed
     • Google Scholar
143. 1. Rammal R
     2. Toulouse G
     3. Virasoro MA

     (1986) Ultrametricity for physicists

     Reviews of Modern Physics 58:765–788.

     https://doi.org/10.1103/RevModPhys.58.765

     • Google Scholar
144. 1. Recanatesi S
     2. Pereira-Obilinovic U
     3. Murakami M
     4. Mainen Z
     5. Mazzucato L

     (2022) Metastable attractors explain the variable timing of stable behavioral action sequences

     Neuron 110:139–153.

     https://doi.org/10.1016/j.neuron.2021.10.011

     • PubMed
     • Google Scholar
145. 1. Reddy G
     2. Desban L
     3. Tanaka H
     4. Roussel J
     5. Mirat O
     6. Wyart C
     7. Faisal AA

     (2022) A lexical approach for identifying behavioural action sequences

     PLOS Computational Biology 18:e1009672.

     https://doi.org/10.1371/journal.pcbi.1009672

     • PubMed
     • Google Scholar
146. 1. Rigotti M
     2. Barak O
     3. Warden MR
     4. Wang XJ
     5. Daw ND
     6. Miller EK
     7. Fusi S

     (2013) The importance of mixed selectivity in complex cognitive tasks

     Nature 497:585–590.

     https://doi.org/10.1038/nature12160

     • PubMed
     • Google Scholar
147. Preprint

     1. Rostami V
     2. Rost T
     3. Riehle A
     4. Albada SJ
     5. Nawrot MP

     (2020) Spiking Neural Network Model of Motor Cortex with Joint Excitatory and Inhibitory Clusters Reflects Task Uncertainty, Reaction Times, and Variability Dynamics

     bioRxiv.

     https://doi.org/10.1101/2020.02.27.968339

     • Google Scholar
148. 1. Roussel MR
     2. Lloyd D

     (2007) Observation of a chaotic multioscillatory metabolic attractor by real-time monitoring of a yeast continuous culture

     The FEBS Journal 274:1011–1018.

     https://doi.org/10.1111/j.1742-4658.2007.05651.x

     • PubMed
     • Google Scholar
149. 1. Samuelsen CL
     2. Gardner MPH
     3. Fontanini A

     (2012) Effects of cue-triggered expectation on cortical processing of taste

     Neuron 74:410–422.

     https://doi.org/10.1016/j.neuron.2012.02.031

     • PubMed
     • Google Scholar
150. 1. Schaub MT
     2. Billeh YN
     3. Anastassiou CA
     4. Koch C
     5. Barahona M

     (2015) Emergence of Slow-Switching Assemblies in Structured Neuronal Networks

     PLOS Computational Biology 11:e1004196.

     https://doi.org/10.1371/journal.pcbi.1004196

     • PubMed
     • Google Scholar
151. 1. Schibler U
     2. Gotic I
     3. Saini C
     4. Gos P
     5. Curie T
     6. Emmenegger Y
     7. Sinturel F
     8. Gosselin P
     9. Gerber A
     10. Fleury-Olela F
     11. Rando G
     12. Demarque M
     13. Franken P

     (2015) Clock-Talk: Interactions between Central and Peripheral Circadian Oscillators in Mammals

     Cold Spring Harbor Symposia on Quantitative Biology 80:223–232.

     https://doi.org/10.1101/sqb.2015.80.027490

     • PubMed
     • Google Scholar
152. Preprint

     1. Schreiner DC
     2. Cazares C
     3. Renteria R
     4. Gremel CM

     (2021) Mice Are Not Automatons; Subjective Experience in Premotor Circuits Guides Behavior

     Neuroscience.

     https://doi.org/10.1101/2021.06.23.449617

     • Google Scholar
153. 1. Schwarz RF
     2. Branicky R
     3. Grundy LJ
     4. Schafer WR
     5. Brown AEX

     (2015) Changes in Postural Syntax Characterize Sensory Modulation and Natural Variation of C. elegans Locomotion

     PLOS Computational Biology 11:e1004322.

     https://doi.org/10.1371/journal.pcbi.1004322

     • PubMed
     • Google Scholar
154. Preprint

     1. Segalin C
     2. Williams J
     3. Karigo T
     4. Hui M
     5. Zelikowsky M
     6. Sun JJ
     7. Perona P
     8. Anderson DJ
     9. Kennedy A

     (2020) The Mouse Action Recognition System (MARS): A Software Pipeline for Automated Analysis of Social Behaviors in Mice

     Neuroscience.

     https://doi.org/10.1101/2020.07.26.222299

     • Google Scholar
155. 1. Shadlen MN
     2. Newsome WT

     (1998)

     The variable discharge of cortical neurons: implications for connectivity, computation, and information coding

     The Journal of Neuroscience 18:3870–3896.

     • PubMed
     • Google Scholar
156. 1. Shi YL
     2. Steinmetz NA
     3. Moore T
     4. Boahen K
     5. Engel TA

     (2022) Cortical state dynamics and selective attention define the spatial pattern of correlated variability in neocortex

     Nature Communications 13:1–15.

     https://doi.org/10.1038/s41467-021-27724-4

     • Google Scholar
157. 1. Simon HA

     (1962)

     The architecture of complexity

     PNAS 106:467.

     • Google Scholar
158. 1. Softky WR
     2. Koch C

     (1993)

     The highly irregular firing of cortical cells is inconsistent with temporal integration of random EPSPs

     The Journal of Neuroscience 13:334–350.

     • PubMed
     • Google Scholar
159. 1. Song S
     2. Sjöström PJ
     3. Reigl M
     4. Nelson S
     5. Chklovskii DB
     6. Friston KJ

     (2005) Highly Nonrandom Features of Synaptic Connectivity in Local Cortical Circuits

     PLOS Biology 3:e68.

     https://doi.org/10.1371/journal.pbio.0030068

     • Google Scholar
160. 1. Sowell ER
     2. Thompson PM
     3. Toga AW

     (2004) Mapping changes in the human cortex throughout the span of life

     The Neuroscientist 10:372–392.

     https://doi.org/10.1177/1073858404263960

     • PubMed
     • Google Scholar
161. 1. Stephens GJ
     2. Johnson-Kerner B
     3. Bialek W
     4. Ryu WS

     (2008) Dimensionality and dynamics in the behavior of C. elegans

     PLOS Computational Biology 4:e1000028.

     https://doi.org/10.1371/journal.pcbi.1000028

     • PubMed
     • Google Scholar
162. Preprint

     1. Stern M
     2. Istrate N
     3. Mazzucato L

     (2021) A Reservoir of Timescales in Random Neural Networks

     bioRxiv.

     https://doi.org/10.1101/2021.10.11.463861

     • Google Scholar
163. 1. Stroud JP
     2. Porter MA
     3. Hennequin G
     4. Vogels TP

     (2018) Motor primitives in space and time via targeted gain modulation in cortical networks

     Nature Neuroscience 21:1774–1783.

     https://doi.org/10.1038/s41593-018-0276-0

     • PubMed
     • Google Scholar
164. Book

     1. Sutton RS
     2. Barto AG.

     (2018)

     Reinforcement learning: An introduction

     MIT press.

     • Google Scholar
165. 1. Szigeti B
     2. Deogade A
     3. Webb B

     (2015) Searching for motifs in the behaviour of larval Drosophila melanogaster and Caenorhabditis elegans reveals continuity between behavioural states

     Journal of the Royal Society, Interface 12:20150899.

     https://doi.org/10.1098/rsif.2015.0899

     • PubMed
     • Google Scholar
166. 1. Tao L
     2. Ozarkar S
     3. Beck JM
     4. Bhandawat V

     (2019) Statistical structure of locomotion and its modulation by odors

     eLife 8:e41235.

     https://doi.org/10.7554/eLife.41235

     • PubMed
     • Google Scholar
167. Book

     1. Tinbergen N

     (1951)

     The Study of Instinct

     Plunkett Lake Press.

     • Google Scholar
168. Preprint

     1. Tishby N
     2. Pereira FC
     3. Bialek W

     (2000) The Information Bottleneck Method

     arXiv.

     https://arxiv.org/abs/physics/0004057

     • Google Scholar
169. 1. Toyoizumi T
     2. Abbott LF

     (2011) Beyond the edge of chaos: amplification and temporal integration by recurrent networks in the chaotic regime

     Physical Review. E, Statistical, Nonlinear, and Soft Matter Physics 84:051908.

     https://doi.org/10.1103/PhysRevE.84.051908

     • PubMed
     • Google Scholar
170. 1. van Beers RJ
     2. Haggard P
     3. Wolpert DM

     (2004) The role of execution noise in movement variability

     Journal of Neurophysiology 91:1050–1063.

     https://doi.org/10.1152/jn.00652.2003

     • PubMed
     • Google Scholar
171. 1. van Daal RJJ
     2. Aydin Ç
     3. Michon F
     4. Aarts AAA
     5. Kraft M
     6. Kloosterman F
     7. Haesler S

     (2021) Implantation of Neuropixels probes for chronic recording of neuronal activity in freely behaving mice and rats

     Nature Protocols 16:3322–3347.

     https://doi.org/10.1038/s41596-021-00539-9

     • PubMed
     • Google Scholar
172. 1. van Vreeswijk C
     2. Sompolinsky H

     (1996) Chaos in neuronal networks with balanced excitatory and inhibitory activity

     Science (New York, N.Y.) 274:1724–1726.

     https://doi.org/10.1126/science.274.5293.1724

     • PubMed
     • Google Scholar
173. 1. Vogelstein JT
     2. Park Y
     3. Ohyama T
     4. Kerr RA
     5. Truman JW
     6. Priebe CE
     7. Zlatic M

     (2014) Discovery of brainwide neural-behavioral maps via multiscale unsupervised structure learning

     Science (New York, N.Y.) 344:386–392.

     https://doi.org/10.1126/science.1250298

     • PubMed
     • Google Scholar
174. 1. Wang J

     (2015) Landscape and flux theory of non-equilibrium dynamical systems with application to biology

     Advances in Physics 64:1–137.

     https://doi.org/10.1080/00018732.2015.1037068

     • Google Scholar
175. 1. Williams AH
     2. Kim TH
     3. Wang F
     4. Vyas S
     5. Ryu SI
     6. Shenoy KV
     7. Schnitzer M
     8. Kolda TG
     9. Ganguli S

     (2018) Unsupervised Discovery of Demixed, Low-Dimensional Neural Dynamics across Multiple Timescales through Tensor Component Analysis

     Neuron 98:1099–1115.

     https://doi.org/10.1016/j.neuron.2018.05.015

     • PubMed
     • Google Scholar
176. 1. Wilson FAW
     2. Scalaidhe SPÓ
     3. Goldman-Rakic PS

     (1993) Dissociation of Object and Spatial Processing Domains in Primate Prefrontal Cortex

     Science 260:1955–1958.

     https://doi.org/10.1126/science.8316836

     • PubMed
     • Google Scholar
177. 1. Wiltschko AB
     2. Johnson MJ
     3. Iurilli G
     4. Peterson RE
     5. Katon JM
     6. Pashkovski SL
     7. Abraira VE
     8. Adams RP
     9. Datta SR

     (2015) Mapping Sub-Second Structure in Mouse Behavior

     Neuron 88:1121–1135.

     https://doi.org/10.1016/j.neuron.2015.11.031

     • PubMed
     • Google Scholar
178. 1. Wiltschko AB
     2. Tsukahara T
     3. Zeine A
     4. Anyoha R
     5. Gillis WF
     6. Markowitz JE
     7. Peterson RE
     8. Katon J
     9. Johnson MJ
     10. Datta SR

     (2020) Revealing the structure of pharmacobehavioral space through motion sequencing

     Nature Neuroscience 23:1433–1443.

     https://doi.org/10.1038/s41593-020-00706-3

     • PubMed
     • Google Scholar
179. 1. Wu HG
     2. Miyamoto YR
     3. Gonzalez Castro LN
     4. Ölveczky BP
     5. Smith MA

     (2014) Temporal structure of motor variability is dynamically regulated and predicts motor learning ability

     Nature Neuroscience 17:312–321.

     https://doi.org/10.1038/nn.3616

     • PubMed
     • Google Scholar
180. 1. Wyrick D
     2. Mazzucato L

     (2021) State-Dependent Regulation of Cortical Processing Speed via Gain Modulation

     The Journal of Neuroscience 41:3988–4005.

     https://doi.org/10.1523/JNEUROSCI.1895-20.2021

     • PubMed
     • Google Scholar
181. 1. Yao Z
     2. Shafer OT

     (2014) The Drosophila circadian clock is a variably coupled network of multiple peptidergic units

     Science (New York, N.Y.) 343:1516–1520.

     https://doi.org/10.1126/science.1251285

     • PubMed
     • Google Scholar
182. 1. Yeatman JD
     2. Wandell BA
     3. Mezer AA

     (2014) Lifespan maturation and degeneration of human brain white matter

     Nature Communications 5:1–12.

     https://doi.org/10.1038/ncomms5932

     • Google Scholar
183. 1. Yu BM
     2. Cunningham JP
     3. Santhanam G
     4. Ryu S
     5. Shenoy KV
     6. Sahani M

     (2008) Gaussian-process factor analysis for low-dimensional single-trial analysis of neural population activity

     Advances in Neural Information Processing Systems 102:614–635.

     https://doi.org/10.1152/jn.90941.2008

     • Google Scholar

Author details

1. Luca Mazzucato

   Institute of Neuroscience, Departments of Biology, Mathematics and Physics, University of Oregon, Eugene, United States

   Contribution

   Conceptualization, Writing - original draft, Writing - review and editing

   For correspondence

   lmazzuca@uoregon.edu

   Competing interests

   No competing interests declared

   "This ORCID iD identifies the author of this article:" 0000-0002-8525-7539

• 2,635

  views

• 447

  downloads

• 21

  citations

Views, downloads and citations are aggregated across all versions of this paper published by eLife.