Contribution of behavioural variability to representational drift

Centering cognitive neuroscience on task demands and generalization

Cognitive neuroscience seeks generalizable theories explaining the relationship between behavioral, physiological and mental states. In pursuit of such theories, we propose a theoretical and empirical framework that centers on understanding task demands and the mutual constraints they impose on behavior and neural activity. Task demands emerge from the interaction between an agent's sensory impressions, goals and behavior, which jointly shape the activity and structure of the nervous system on multiple spatiotemporal scales. Understanding this interaction requires multitask studies that vary more than one experimental component (for example, stimuli and instructions) combined with dense behavioral and neural sampling and explicit testing for generalization across tasks and data modalities. By centering task demands rather than mental processes that tasks are assumed to engage, this framework paves the way for the discovery of new generalizable concepts unconstrained by existing taxonomies, and moves cognitive neuroscience toward an action-oriented, dynamic and integrated view of the brain.

Drift of neural ensembles driven by slow fluctuations of intrinsic excitability

Representational drift refers to the dynamic nature of neural representations in the brain despite the behavior being seemingly stable. Although drift has been observed in many different brain regions, the mechanisms underlying it are not known. Since intrinsic neural excitability is suggested to play a key role in regulating memory allocation, fluctuations of excitability could bias the reactivation of previously stored memory ensembles and therefore act as a motor for drift. Here, we propose a rate-based plastic recurrent neural network with slow fluctuations of intrinsic excitability. We first show that subsequent reactivations of a neural ensemble can lead to drift of this ensemble. The model predicts that drift is induced by co-activation of previously active neurons along with neurons with high excitability which leads to remodeling of the recurrent weights. Consistent with previous experimental works, the drifting ensemble is informative about its temporal history. Crucially, we show that the gradual nature of the drift is necessary for decoding temporal information from the activity of the ensemble. Finally, we show that the memory is preserved and can be decoded by an output neuron having plastic synapses with the main region.

Representational drift as a result of implicit regularization

Recent studies show that, even in constant environments, the tuning of single neurons changes over time in a variety of brain regions. This representational drift has been suggested to be a consequence of continuous learning under noise, but its properties are still not fully understood. To investigate the underlying mechanism, we trained an artificial network on a simplified navigational task. The network quickly reached a state of high performance, and many units exhibited spatial tuning. We then continued training the network and noticed that the activity became sparser with time. Initial learning was orders of magnitude faster than ensuing sparsification. This sparsification is consistent with recent results in machine learning, in which networks slowly move within their solution space until they reach a flat area of the loss function. We analyzed four datasets from different labs, all demonstrating that CA1 neurons become sparser and more spatially informative with exposure to the same environment. We conclude that learning is divided into three overlapping phases: (i) Fast familiarity with the environment; (ii) slow implicit regularization; and (iii) a steady state of null drift. The variability in drift dynamics opens the possibility of inferring learning algorithms from observations of drift statistics.

Perpetual step-like restructuring of hippocampal circuit dynamics

Representation of the environment by hippocampal populations is known to drift even within a familiar environment, which could reflect gradual changes in single cell activity or result from averaging across discrete switches of single neurons. Disambiguating these possibilities is crucial, as they each imply distinct mechanisms. Leveraging change point detection and model comparison, we found that CA1 population vectors decorrelated gradually within a session. In contrast, individual neurons exhibited predominantly step-like emergence and disappearance of place fields or sustained change in within-field firing. The changes were not restricted to particular parts of the maze or trials and did not require apparent behavioral changes. The same place fields emerged, disappeared, and reappeared across days, suggesting that the hippocampus reuses pre-existing assemblies, rather than forming new fields de novo. Our results suggest an internally-driven perpetual step-like reorganization of the neuronal assemblies.

Spontaneous Dynamics of Hippocampal Place Fields in a Model of Combinatorial Competition among Stable Inputs

We present computer simulations illustrating how the plastic integration of spatially stable inputs could contribute to the dynamic character of hippocampal spatial representations. In novel environments of slightly larger size than typical apparatus, the emergence of well-defined place fields in real place cells seems to rely on inputs from normally functioning grid cells. Theoretically, the grid-to-place transformation is possible if a place cell is able to respond selectively to a combination of suitably aligned grids. We previously identified the functional characteristics that allow a synaptic plasticity rule to accomplish this selection by synaptic competition during rat foraging behavior. Here, we show that the synaptic competition can outlast the formation of place fields, contributing to their spatial reorganization over time, when the model is run in larger environments and the topographical/modular organization of grid inputs is taken into account. Co-simulated cells that differ only by their randomly assigned grid inputs display different degrees and kinds of spatial reorganization-ranging from place-field remapping to more subtle in-field changes or lapses in firing. The model predicts a greater number of place fields and propensity for remapping in place cells recorded from more septal regions of the hippocampus and/or in larger environments, motivating future experimental standardization across studies and animal models. In sum, spontaneous remapping could arise from rapid synaptic learning involving inputs that are functionally homogeneous, spatially stable, and minimally stochastic.

A persistent prefrontal reference frame across time and task rules

Behavior can be remarkably consistent, even over extended time periods, yet whether this is reflected in stable or 'drifting' neuronal responses to task features remains controversial. Here, we find a persistently active ensemble of neurons in the medial prefrontal cortex (mPFC) of mice that reliably maintains trajectory-specific tuning over several weeks while performing an olfaction-guided spatial memory task. This task-specific reference frame is stabilized during learning, upon which repeatedly active neurons show little representational drift and maintain their trajectory-specific tuning across long pauses in task exposure and across repeated changes in cue-target location pairings. These data thus suggest a 'core ensemble' of prefrontal neurons forming a reference frame of task-relevant space for the performance of consistent behavior over extended periods of time.

Representational drift as a result of implicit regularization

Recent studies show that, even in constant environments, the tuning of single neurons changes over time in a variety of brain regions. This representational drift has been suggested to be a consequence of continuous learning under noise, but its properties are still not fully understood. To investigate the underlying mechanism, we trained an artificial network on a simplified navigational task. The network quickly reached a state of high performance, and many units exhibited spatial tuning. We then continued training the network and noticed that the activity became sparser with time. Initial learning was orders of magnitude faster than ensuing sparsification. This sparsification is consistent with recent results in machine learning, in which networks slowly move within their solution space until they reach a flat area of the loss function. We analyzed four datasets from different labs, all demonstrating that CA1 neurons become sparser and more spatially informative with exposure to the same environment. We conclude that learning is divided into three overlapping phases: (i) Fast familiarity with the environment; (ii) slow implicit regularization; (iii) a steady state of null drift. The variability in drift dynamics opens the possibility of inferring learning algorithms from observations of drift statistics.

Hippocampal sequences span experience relative to rewards

Hippocampal place cells fire in sequences that span spatial environments and non-spatial modalities, suggesting that hippocampal activity can anchor to the most behaviorally salient aspects of experience. As reward is a highly salient event, we hypothesized that sequences of hippocampal activity can anchor to rewards. To test this, we performed two-photon imaging of hippocampal CA1 neurons as mice navigated virtual environments with changing hidden reward locations. When the reward moved, the firing fields of a subpopulation of cells moved to the same relative position with respect to reward, constructing a sequence of reward-relative cells that spanned the entire task structure. The density of these reward-relative sequences increased with task experience as additional neurons were recruited to the reward-relative population. Conversely, a largely separate subpopulation maintained a spatially-based place code. These findings thus reveal separate hippocampal ensembles can flexibly encode multiple behaviorally salient reference frames, reflecting the structure of the experience.

Unconscious Manipulation of Conceptual Representations with Decoded Neurofeedback Impacts Search Behavior

The necessity of conscious awareness in human learning has been a long-standing topic in psychology and neuroscience. Previous research on non-conscious associative learning is limited by the low signal-to-noise ratio of the subliminal stimulus, and the evidence remains controversial, including failures to replicate. Using functional MRI decoded neurofeedback, we guided participants from both sexes to generate neural patterns akin to those observed when visually perceiving real-world entities (e.g., dogs). Importantly, participants remained unaware of the actual content represented by these patterns. We utilized an associative DecNef approach to imbue perceptual meaning (e.g., dogs) into Japanese hiragana characters that held no inherent meaning for our participants, bypassing a conscious link between the characters and the dogs concept. Despite their lack of awareness regarding the neurofeedback objective, participants successfully learned to activate the target perceptual representations in the bilateral fusiform. The behavioral significance of our training was evaluated in a visual search task. DecNef and control participants searched for dogs or scissors targets that were pre-cued by the hiragana used during DecNef training or by a control hiragana. The DecNef hiragana did not prime search for its associated target but, strikingly, participants were impaired at searching for the targeted perceptual category. Hence, conscious awareness may function to support higher-order associative learning. Meanwhile, lower-level forms of re-learning, modification, or plasticity in existing neural representations can occur unconsciously, with behavioral consequences outside the original training context. The work also provides an account of DecNef effects in terms of neural representational drift.

Reward Expectation Reduces Representational Drift in the Hippocampus

Spatial memory in the hippocampus involves dynamic neural patterns that change over days, termed representational drift. While drift may aid memory updating, excessive drift could impede retrieval. Memory retrieval is influenced by reward expectation during encoding, so we hypothesized that diminished reward expectation would exacerbate representational drift. We found that high reward expectation limited drift, with CA1 representations on one day gradually re-emerging over successive trials the following day. Conversely, the absence of reward expectation resulted in increased drift, as the gradual re-emergence of the previous day's representation did not occur. At the single cell level, lowering reward expectation caused an immediate increase in the proportion of place-fields with low trial-to-trial reliability. These place fields were less likely to be reinstated the following day, underlying increased drift in this condition. In conclusion, heightened reward expectation improves memory encoding and retrieval by maintaining reliable place fields that are gradually reinstated across days, thereby minimizing representational drift.

Arousal as a universal embedding for spatiotemporal brain dynamics

Neural activity in awake organisms shows widespread and spatiotemporally diverse correlations with behavioral and physiological measurements. We propose that this covariation reflects in part the dynamics of a unified, arousal-related process that regulates brain-wide physiology on the timescale of seconds. Taken together with theoretical foundations in dynamical systems, this interpretation leads us to a surprising prediction: that a single, scalar measurement of arousal (e.g., pupil diameter) should suffice to reconstruct the continuous evolution of multimodal, spatiotemporal measurements of large-scale brain physiology. To test this hypothesis, we perform multimodal, cortex-wide optical imaging and behavioral monitoring in awake mice. We demonstrate that spatiotemporal measurements of neuronal calcium, metabolism, and blood-oxygen can be accurately and parsimoniously modeled from a low-dimensional state-space reconstructed from the time history of pupil diameter. Extending this framework to behavioral and electrophysiological measurements from the Allen Brain Observatory, we demonstrate the ability to integrate diverse experimental data into a unified generative model via mappings from an intrinsic arousal manifold. Our results support the hypothesis that spontaneous, spatially structured fluctuations in brain-wide physiology-widely interpreted to reflect regionally-specific neural communication-are in large part reflections of an arousal-related process. This enriched view of arousal dynamics has broad implications for interpreting observations of brain, body, and behavior as measured across modalities, contexts, and scales.

Reconstructing computational system dynamics from neural data with recurrent neural networks

Computational models in neuroscience usually take the form of systems of differential equations. The behaviour of such systems is the subject of dynamical systems theory. Dynamical systems theory provides a powerful mathematical toolbox for analysing neurobiological processes and has been a mainstay of computational neuroscience for decades. Recently, recurrent neural networks (RNNs) have become a popular machine learning tool for studying the non-linear dynamics of neural and behavioural processes by emulating an underlying system of differential equations. RNNs have been routinely trained on similar behavioural tasks to those used for animal subjects to generate hypotheses about the underlying computational mechanisms. By contrast, RNNs can also be trained on the measured physiological and behavioural data, thereby directly inheriting their temporal and geometrical properties. In this way they become a formal surrogate for the experimentally probed system that can be further analysed, perturbed and simulated. This powerful approach is called dynamical system reconstruction. In this Perspective, we focus on recent trends in artificial intelligence and machine learning in this exciting and rapidly expanding field, which may be less well known in neuroscience. We discuss formal prerequisites, different model architectures and training approaches for RNN-based dynamical system reconstructions, ways to evaluate and validate model performance, how to interpret trained models in a neuroscience context, and current challenges.

Mega-scale movie-fields in the mouse visuo-hippocampal network

Natural visual experience involves a continuous series of related images while the subject is immobile. How does the cortico-hippocampal circuit process a visual episode? The hippocampus is crucial for episodic memory, but most rodent single unit studies require spatial exploration or active engagement. Hence, we investigated neural responses to a silent movie (Allen Brain Observatory) in head-fixed mice without any task or locomotion demands, or rewards. Surprisingly, a third (33%, 3379/10263) of hippocampal -dentate gyrus, CA3, CA1 and subiculum- neurons showed movie-selectivity, with elevated firing in specific movie sub-segments, termed movie-fields, similar to the vast majority of thalamo-cortical (LGN, V1, AM-PM) neurons (97%, 6554/6785). Movie-tuning remained intact in immobile or spontaneously running mice. Visual neurons had >5 movie-fields per cell, but only ~2 in hippocampus. The movie-field durations in all brain regions spanned an unprecedented 1000-fold range: from 0.02s to 20s, termed mega-scale coding. Yet, the total duration of all the movie-fields of a cell was comparable across neurons and brain regions. The hippocampal responses thus showed greater continuous-sequence encoding than visual areas, as evidenced by fewer and broader movie-fields than in visual areas. Consistently, repeated presentation of the movie images in a fixed, but scrambled sequence virtually abolished hippocampal but not visual-cortical selectivity. The preference for continuous, compared to scrambled sequence was eight-fold greater in hippocampal than visual areas, further supporting episodic-sequence encoding. Movies could thus provide a unified way to probe neural mechanisms of episodic information processing and memory, even in immobile subjects, across brain regions, and species.

Computational assessment of visual coding across mouse brain areas and behavioural states

Introduction

Our brain is bombarded by a diverse range of visual stimuli, which are converted into corresponding neuronal responses and processed throughout the visual system. The neural activity patterns that result from these external stimuli vary depending on the object or scene being observed, but they also change as a result of internal or behavioural states. This raises the question of to what extent it is possible to predict the presented visual stimuli from neural activity across behavioural states, and how this varies in different brain regions.

Methods

To address this question, we assessed the computational capacity of decoders to extract visual information in awake behaving mice, by analysing publicly available standardised datasets from the Allen Brain Institute. We evaluated how natural movie frames can be distinguished based on the activity of units recorded in distinct brain regions and under different behavioural states. This analysis revealed the spectrum of visual information present in different brain regions in response to binary and multiclass classification tasks.

Results

Visual cortical areas showed highest classification accuracies, followed by thalamic and midbrain regions, with hippocampal regions showing close to chance accuracy. In addition, we found that behavioural variability led to a decrease in decoding accuracy, whereby large behavioural changes between train and test sessions reduced the classification performance of the decoders. A generalised linear model analysis suggested that this deterioration in classification might be due to an independent modulation of neural activity by stimulus and behaviour. Finally, we reconstructed the natural movie frames from optimal linear classifiers, and observed a strong similarity between reconstructed and actual movie frames. However, the similarity was significantly higher when the decoders were trained and tested on sessions with similar behavioural states.

Conclusion

Our analysis provides a systematic assessment of visual coding in the mouse brain, and sheds light on the spectrum of visual information present across brain areas and behavioural states.

Time and experience are independent determinants of representational drift in CA1

In this issue of Neuron, Khatib et al.1 and Geva et al.2 present complementary and breakthrough discoveries demonstrating that elapsed time and active experience independently affect unique aspects of representational drift in the hippocampus.

Representational drift as a window into neural and behavioural plasticity

Large-scale recordings of neural activity over days and weeks have revealed that neural representations of familiar tasks, precepts and actions continually evolve without obvious changes in behaviour. We hypothesise that this steady drift in neural activity and accompanying physiological changes is due in part to the continuous application of a learning rule at the cellular and population level. Explicit predictions of this drift can be found in neural network models that use iterative learning to optimise weights. Drift therefore provides a measurable signal that can reveal systems-level properties of biological plasticity mechanisms, such as their precision and effective learning rates.

Representations in human primary visual cortex drift over time

Primary sensory regions are believed to instantiate stable neural representations, yet a number of recent rodent studies suggest instead that representations drift over time. To test whether sensory representations are stable in human visual cortex, we analyzed a large longitudinal dataset of fMRI responses to images of natural scenes. We fit the fMRI responses using an image-computable encoding model and tested how well the model generalized across sessions. We found systematic changes in model fits that exhibited cumulative drift over many months. Convergent analyses pinpoint changes in neural responsivity as the source of the drift, while population-level representational dissimilarities between visual stimuli were unchanged. These observations suggest that downstream cortical areas may read-out a stable representation, even as representations within V1 exhibit drift.

Time and experience differentially affect distinct aspects of hippocampal representational drift

Hippocampal activity is critical for spatial memory. Within a fixed, familiar environment, hippocampal codes gradually change over timescales of days to weeks-a phenomenon known as representational drift. The passage of time and the amount of experience are two factors that profoundly affect memory. However, thus far, it has remained unclear to what extent these factors drive hippocampal representational drift. Here, we longitudinally recorded large populations of hippocampal neurons in mice while they repeatedly explored two different familiar environments that they visited at different time intervals over weeks. We found that time and experience differentially affected distinct aspects of representational drift: the passage of time drove changes in neuronal activity rates, whereas experience drove changes in the cells' spatial tuning. Changes in spatial tuning were context specific and largely independent of changes in activity rates. Thus, our results suggest that representational drift is a multi-faceted process governed by distinct neuronal mechanisms.

Long-term stability of single neuron activity in the motor system

How an established behavior is retained and consistently produced by a nervous system in constant flux remains a mystery. One possible solution to ensure long-term stability in motor output is to fix the activity patterns of single neurons in the relevant circuits. Alternatively, activity in single cells could drift over time provided that the population dynamics are constrained to produce the same behavior. To arbitrate between these possibilities, we recorded single-unit activity in motor cortex and striatum continuously for several weeks as rats performed stereotyped motor behaviors-both learned and innate. We found long-term stability in single neuron activity patterns across both brain regions. A small amount of drift in neural activity, observed over weeks of recording, could be explained by concomitant changes in task-irrelevant aspects of the behavior. These results suggest that long-term stable behaviors are generated by single neuron activity patterns that are themselves highly stable.

The geometry of representational drift in natural and artificial neural networks

Neurons in sensory areas encode/represent stimuli. Surprisingly, recent studies have suggested that, even during persistent performance, these representations are not stable and change over the course of days and weeks. We examine stimulus representations from fluorescence recordings across hundreds of neurons in the visual cortex using in vivo two-photon calcium imaging and we corroborate previous studies finding that such representations change as experimental trials are repeated across days. This phenomenon has been termed "representational drift". In this study we geometrically characterize the properties of representational drift in the primary visual cortex of mice in two open datasets from the Allen Institute and propose a potential mechanism behind such drift. We observe representational drift both for passively presented stimuli, as well as for stimuli which are behaviorally relevant. Across experiments, the drift differs from in-session variance and most often occurs along directions that have the most in-class variance, leading to a significant turnover in the neurons used for a given representation. Interestingly, despite this significant change due to drift, linear classifiers trained to distinguish neuronal representations show little to no degradation in performance across days. The features we observe in the neural data are similar to properties of artificial neural networks where representations are updated by continual learning in the presence of dropout, i.e. a random masking of nodes/weights, but not other types of noise. Therefore, we conclude that a potential reason for the representational drift in biological networks is driven by an underlying dropout-like noise while continuously learning and that such a mechanism may be computational advantageous for the brain in the same way it is for artificial neural networks, e.g. preventing overfitting.

Visual experience has opposing influences on the quality of stimulus representation in adult primary visual cortex

Transient dark exposure, typically 7–10 days in duration, followed by light reintroduction is an emerging treatment for improving the restoration of vision in amblyopic subjects whose occlusion is removed in adulthood. Dark exposure initiates homeostatic mechanisms that together with light-induced changes in cellular signaling pathways result in the re-engagement of juvenile-like plasticity in the adult such that previously deprived inputs can gain cortical territory. It is possible that dark exposure itself degrades visual responses, and this could place constraints on the optimal duration of dark exposure treatment. To determine whether eight days of dark exposure has a lasting negative impact on responses to classic grating stimuli, neural activity was recorded before and after dark exposure in awake head-fixed mice using two-photon calcium imaging. Neural discriminability, assessed using classifiers, was transiently reduced following dark exposure; a decrease in response reliability across a broad range of spatial frequencies likely contributed to the disruption. Both discriminability and reliability recovered. Fixed classifiers were used to demonstrate that stimulus representation rebounded to the original, pre-deprivation state, thus dark exposure did not appear to have a lasting negative impact on visual processing. Unexpectedly, we found that dark exposure significantly stabilized orientation preference and signal correlation. Our results reveal that natural vision exerts a disrupting influence on the stability of stimulus preference for classic grating stimuli and, at the same time, improves neural discriminability for both low and high-spatial frequency stimuli.