Biology subjects

Gershman, S. J.

Publications and source records attributed to Gershman, S. J..

At least 19 recordsLinked to original sources

bioRxiv2026

Dynamic belief representation and updating through learned attractor-like dynamics in the frontal cortex

To act adaptively, animals must infer hidden states of the world from incomplete sensory information and update beliefs as new observations accrue. While dopamine signals are well explained by reinforcement learning models that incorporate belief states, how the brain implements belief-state inference remains unknown. Prior modeling work showed that recurrent neural networks trained to predict value (Value-RNN) develop task-specific, attractor-like dynamics that mirror evolution of beliefs. Here we performed high-density electrophysiological recordings from orbitofrontal cortex (OFC) and other brain areas of mice performing two variants of a Pavlovian task that differ in reward probability, which produce different within-trial belief dynamics. We find that OFC population activity exhibits task-specific attractor-like dynamics that mirror the within-trial dynamics in the Value-RNN. These dynamics are absent in motor and olfactory regions, are not explained by behavioral differences, and emerge progressively with learning. Our findings indicate that the brain approximates belief-state inference through learned, task-specific attractor-like dynamics.

neuroscience↗

bioRxiv2026

Reliability-guided meta-control in intuitive physics

Intuitive physical reasoning is an important part of daily life, but the computations underlying it remain debated. Some prominent accounts propose intuitive physics relies on mental simulation, with people evolving mental scenes forward through step-by-step. However, representing every intermediate mental state in detail can be computationally costly. We suggest that intuitive physical reasoning combines mental simulation with learned predictive abstractions, and that the trade-off between these two is determined by the prospective reliability of an abstraction. We implemented the abstraction part of our proposal as a recurrent neural network model that learns a predictive abstraction by directly predicting the object's future position from a short visual history. This network bypasses intermediate states, while also estimating its own prediction error. Our network produced simplified linear trajectories that shortcut ground-truth physical trajectories (in line with human behavior), and its error estimates were associated with human pupil size during a physical prediction task. We further developed a meta-control model that used the network's error estimate to arbitrate between abstraction (direct prediction by the network) and simulation (running a mental simulation). The meta-control model captures key patterns in human response time and accuracy, and its predicted abstraction and simulation phases corresponded to distinct eye-movement signatures, over and above an earlier blended model that uses a hand-designed abstraction. Our results suggest that efficient physical reasoning can emerge from an interaction between learned abstractions, estimates of how reliable these abstractions are, and selective engagement of detailed simulation based on reliability estimates.

animal behavior and cognition↗

bioRxiv2026

The Reuse-and-Append Memory Principle: Application to Latent Cause Inference

The latent cause theory of memory modification provides a computational account of how the brain decides whether to update existing memories or form new ones, but leaves unspecified the neural mechanisms implementing this inference. We propose Reuse-and-Append Memory (RAM), a set of mechanistic principles that achieve the same goal through sparse neural coding, Hebbian learning, and pattern-matching dynamics. The core idea is that neurons encoding prior experiences are automatically reactivated by similar stimuli, while uncommitted neurons are recruited to encode genuinely novel aspects of each experience, including the passage of time. We present a computational model instantiating these principles and show that it reproduces acquisition, extinction, renewal, and spontaneous recovery in fear conditioning. By committing to a neural mechanism, we found that RAM separates what is typically modeled as a single prediction error driving memory updating or formation into two independent signals: an immediate novelty signal that emerges from coverage-based allocation of new neurons, and an outcome mismatch signal that later activates safety circuits when an expected outcome fails to arrive. RAM also reproduces the dependence of recovery on the timing of stimulus reminders (the Monfils-Schiller effect), but predicts it to be inherently fragile, consistent with the mixed empirical record. RAM suggests that the Monfils-Schiller effect, often attributed to reconsolidation, leaves the original fear memory intact: a bridging event carries safety information to new contexts through offline co-retrieval. More broadly, none of the phenomena we explain require modifying existing memories, and the most stable configuration is the extreme form of the principle: always append, never overwrite.

neuroscience↗

bioRxiv2026

Policy regularization as a unifying theory of the striatal division of labor in learning

Learning novel behaviors requires balancing previously learned actions with the ability to flexibly adapt to changing reward contingencies. This trade-off is well documented in the division of labor between dorsolateral striatum (DLS), which promotes selection of cached, history-dependent actions, and dorsomedial striatum (DMS), which supports flexible learning as reward contingencies change. Here, we propose policy regularization as a general computational principle for understanding this division. Capacity-limited agents face a fundamental trade-off between maximizing reward and minimizing the cost of deviating from a default policy that caches frequently used action transitions. We formalize this trade-off as a KL-regularized reward objective in which a flexible controller (DMS) incurs a cost for diverging from a history-dependent default (DLS). The resulting optimal policy is a combination of a reward-driven action value, continuously updated by DMS, and a default policy conditioned on action history, cached by DLS. In practice, DLS consolidates the action transitions shaped by DMS's reward-driven value learning, so DLS preserves behaviors that were once optimal even when reward contingencies change, producing robust but inflexible action selection and effectively "regularizing" DMS-driven learning. We show that a single model with one shared parameter set and consistent lesion rules provides a unifying explanation for the functional organization of the striatum, reproducing canonical DLS-DMS dissociations in outcome devaluation, serial spatial reversal, skilled action sequencing, and motor sequence execution tasks.

neuroscience↗

bioRxiv2026

Efficient coding makes and breaks Webers law

Webers law is a rare quantitative regularity in psychology, yet its origins remain debated. Here we provide causal evidence that it arises from the more fundamental principle of efficient coding. This principle posits that representational resources are allocated according to stimulus frequencies: distributions skewed toward smaller stimuli thus result in discriminability decreasing with magnitude, as in Webers law. Skewing frequencies in the other direction--making large magnitudes more frequent than small ones--enabled us to invert this pattern, and to break Webers law. In discrimination tasks with three different sensory modalities, human subjects discriminability across stimuli was sensitive to the stimulus distribution, and this adaptation improved task performance. These findings establish efficient coding as a dynamic, organizing principle, explaining when and why Webers law holds.

neuroscience↗

bioRxiv2026

An improved Stentor coeruleus genome and time-resolved transcriptomics link cyclic nucleotide-dependent kinase signaling to single-cell habituation

The giant ciliate Stentor coeruleus is a single cell capable of modifying its behavior through experience. When repeatedly disturbed, it undergoes habituation, a simple and widely conserved form of non-associative learning, despite lacking a nervous system. How a single cell achieves such behavioral plasticity at the molecular level remains poorly understood. Here, we report a compact, contiguous macronuclear reference genome for S. coeruleus, substantially improving genome continuity and annotation quality while resolving two successive rounds of whole-genome duplication and extensive lineage-specific gene-family expansion. Anchoring time-resolved transcriptomics to this reference, we profiled pooled RNA from behaviorally tracked cells collected at defined points during a mechanical stimulation paradigm and identified 341 genes with significant temporal expression dynamics during habituation. The dominant transcriptional signature of habituation was the coordinated, sustained up-regulation of a conserved cyclic nucleotide, calcium, and phosphatase signaling module centered on five cGMP-dependent protein kinase (PKG) paralogs together with a voltage-gated calcium channel and a protein phosphatase 2A regulatory subunit, representing conserved intracellular components of the neuronal long-term depression (LTD) pathway. By integrating an improved genome with time-resolved transcriptomics of habituation, this work provides a community resource for Stentor and identifies a conserved signaling program as a candidate molecular framework underlying single-cell habituation.

molecular biology↗

bioRxiv2026

A quantitative portrait of habituation in Stentor coeruleus

Habituation--the decrement in response to a series of stimuli--is a widespread form of learning observed across many organisms, including the unicellular organism Stentor coeruleus. A lesser-known feature of Stentor habituation, shared with animals, is potentiation: faster habituation to a second stimulus series despite partial or complete recovery of responsiveness before that series begins. This suggests that although the first-order habituation memory can decay during the recovery period between the two series, a persistent second-order memory mediates faster relearning. We investigate the response profile of Stentor across a range of stimulation frequencies and recovery periods to identify the timescales at which these memory traces operate. We introduce a statistical framework to infer both population and single-cell learning parameters, allowing us to quantify prior qualitative findings and examine relationships among parameters across cells. Two key findings are that potentiation is frequency-sensitive, and that recovery and potentiation are decoupled, consistent with a serial and hierarchical cascade of leaky integrator units underlying these processes. This quantitative portrait provides a foundation for mechanistic modeling of intracellular memory in Stentor.

cell biology↗

bioRxiv2026

Failure to classically condition planarian flatworms

Planarian flatworms represent one of the most evolutionarily informative nervous systems for an account of ancient bilaterian brains. Likewise, the unparalleled regenerative ability of planarians makes possible certain investigations of neural development, memory, and behavior that are simply impossible with other model organisms. Despite these facts, learning and memory are today underexplored in planarians, likely due in part to the controversial legacy of 20th-century experiments on the transfer of memories between individual flatworms. Here, we attempted to replicate and extend the classical conditioning experiments in planarians that were the basis of the later memory transfer work. We failed to find evidence of learning using both historical and contemporary conditioning protocols, and using computer vision methods to avoid subjectivity in manual video annotation. Our results cast doubt on the suitability of historic conditioning protocols for Girardia dorotocephala and Schmidtea mediterranea planarians. We encourage future researchers to explore more ethologically relevant conditioning paradigms to establish a behavioral foundation for investigating the molecular basis of learning.

animal behavior and cognition↗

bioRxiv2026

Phasic dopamine drives conditioned responding beyond its role in learning

Animals exposed to pairings of a neutral stimulus with reward acquire a conditioned response to the neutral stimulus. A prominent hypothesis, formalized in the Temporal Difference (TD) learning algorithm, is that animals learn to predict the future reward associated with the neutral stimulus ("value"). Though the TD algorithm does not explicitly specify what drives conditioned responding, a typical assumption is that it reflects the animals estimate of value. In TD learning, value estimates are updated using reward prediction error (RPE, the discrepancy between observed and predicted reward), and are thought to be signaled by the phasic activity of midbrain dopamine neurons. This hypothesis posits that dopamines effects on conditioned responding are mediated entirely by its effects on learning. However, recent experimental and theoretical evidence suggests that dopamine may play a more direct role in modulating conditioned responding. We use a combination of data analysis and computational modeling to probe the relationship between dopamine and conditioned responding. Our results suggest that dopamine directly modulates conditioned responding, in addition to its role in learning. These findings can be captured by a model in which dopamine RPE acts both indirectly (via learning) and directly on conditioned responding.

neuroscience↗

bioRxiv2026

Associative learning in the protozoan Stentor coeruleus

The capacity for associative learning in protozoa has been a matter of longstanding controversy. In a series of Pavlovian conditioning experiments with the ciliate Stentor coeruleus, we show that temporally pairing weak and strong mechanical stimuli results in a transiently enhanced contraction response to the weak stimulus. Control experiments rule out several alternative explanations, such as non-associative sensitization or arousal. Parametric manipulation of the conditioning protocol's temporal structure revealed a systematic dependence of learning on the inter-trial and inter-stimulus intervals, though not in the form classically observed in animals. A simple mathematical model, combining associative learning with habituation, can explain why enhancement is transient, and accurately fits the learning curve at the aggregate level. We conclude that Stentor coeruleus appears capable of associative learning, suggesting an ancient evolutionary origin that preceded the emergence of multicellular nervous systems.

cell biology↗

bioRxiv2025

Transition from plasticity to dynamics-based value update for meta-learning

The ability to estimate the value associated with a specific stimulus or action is essential for adaptive behavior. Value can be updated either incrementally through experience or rapidly by inference based on latent environmental structure. Yet, how the brain implements and transitions between these modes of value computation remains unclear. To address this question, we examined the neuronal mechanisms underlying reversal learning. Mice were trained in an odor-outcome association task either with stable contingencies or with dynamically changing contingencies. Mice trained on stable contingencies formed long-term value representations that depended on synaptic plasticity in the basolateral amygdala (BLA). In contrast, mice exposed to repeated reversals acquired the ability to infer values, independent from plasticity in BLA, enabling faster learning but with more rapid memory decay. Recurrent neural network models (RNNs) trained with continuous weight updates recapitulated this transition, shifting from plasticity-based to dynamics-based value computation. Neural activity in the BLA encoded both value and contextual information necessary for computing value based on latent task structure, similar to those found in the RNNs. Disrupting BLA activity before cue delivery preferentially impaired dynamics-based value updating. Furthermore, mice could learn distinct correlation structures that enabled structure-specific value inference. Together, these findings provide a mechanistic framework for fast value updates via inference, a core feature of intelligent behavior.

neuroscience↗

bioRxiv2025

Storing long-lived memories via molecular error correction: a minimal mathematical model of Crick's memory switch

Cells store information in part by attaching molecular marks to proteins and DNA. But because marks can be randomly removed (e.g., due to ambient phosphatase activity) and added (e.g., due to ambient kinase activity), information encoding may be poor, and in the worst case marks may only store information on the time scale of molecular turnover. We identify a high-level strategy cells use to maintain encoding fidelity beyond the time scale of turnover-- which we call molecular error correction--and find that it closely resembles an underexplored theoretical proposal by Francis Crick. To assess the effectiveness of molecular error correction, we construct and analyze several minimal mathematical models of molecular memory switches. We find that Crick-like error correction provides an efficient way to improve the lifetime of stored memories, especially compared to redundantly encoding information using a large number of molecules, but that it can yield false positives (and hence low information fidelity) when there is ambient marking activity; given these two competing concerns, the optimal level of error correction is moderate rather than arbitrarily high. We also find that combining error correction with redundant encoding can efficiently and robustly produce memories that last between ten and one hundred times longer than the turnover time scale. Our work provides insight regarding how to model and interrogate noisy molecular memory systems, and suggests that error correction is a design principle of performant molecular memory.

systems biology↗

bioRxiv2025

Distributed range adaptation in human parietal encoding of numbers

The brain's representations are encoded in the collective activity of neural populations, whose non-monotonic tuning properties define the population code. Adaptive behavior calls for flexible retuning of this code as stimuli statistics change across contexts. Yet how such adaptation occurs is unknown. Using fMRI during a numerosity-estimation task with variable number ranges, we show that number representations in human parietal cortex dynamically recalibrate to context, enabling context-sensitive behavioral accuracy. The tuning properties of neural populations collectively shift and rescale with the range. This distributed range adaptation achieves efficient coding in real time: neural precision scales with the range and predicts corresponding changes in behavioral precision. Individuals with stronger neural adaptation show larger behavioral adjustments. These findings extend static sensory efficient coding to adaptive representations of abstract magnitudes. Such flexible population tuning may constitute a canonical mechanism of encoding networks that enables the brain to sustain precise, adaptive behavior.

neuroscience↗

bioRxiv2025

Entorhinal cortex signals dimensions of past experience that can be generalised in a novel environment

No two situations are identical. They can be similar in some aspects but different in others. This poses a key challenge when attempting to generalise our experience from one situation to another. How do we distinguish the aspects that transfer across situations from those that do not? One hypothesis is that the entorhinal cortex (EC) meets this challenge by forming factorised representations that allow for increased neural similarity between events that share generalisable features. We tested this hypothesis using functional magnetic resonance imaging. Female and male participants (n=40) were trained to report behavioural sequences based on an underlying graph structure. Participants then made decisions in a new environment where some, but not all graph transitions from the previous structure could be generalised. Behavioural results showed that participants distinguished the generalisable transition information. Accuracy was significantly higher in blocks where sequence transitions were shared across environments, than those in which transitions differed. This boost in accuracy was especially pronounced during early exposure to the novel environment. Throughout this early phase, neural patterns in EC showed a corresponding differentiation of the generalisable aspects. Neural patterns representing starting locations in familiar and novel environments were significantly more similar in EC on trials where sequences could be generalised from prior experience, compared to trials with new sequential transitions. This signalling was associated with improved performance when prior sequence knowledge could be reused. Our results suggest that during early exposure to novel environments, EC may signal dimensions of past experience that can be generalised. Significance StatementGeneralisation is the process of using our experience to solve new challenges. A central problem we face when attempting to generalise is determining which aspects of a new situation are similar to what we have experienced before and which are different. This allows us to generalise our knowledge selectively, transferring insights to the aspects that are similar and relevant in novel scenarios, while avoiding over-generalisation to the aspects that differ. The results from this experiment suggest that the entorhinal cortex may distinguish these aspects during early exposure to new environments, supporting complex generalisation behaviour.

neuroscience↗

bioRxiv2025

Fast efficient coding and sensory adaptation in gain-adaptive recurrent networks

As the statistics of sensory environments often change, neural sensory systems must adapt to maintain useful representations. Efficient coding prescribes that neuronal tuning curves should be optimized to the prior, but whether they can adapt rapidly is unclear. Empirically, tuning curves after repeated stimulus presentations exhibit "adapter repulsion", whose underlying mechanism remains uncertain, and which contrasts with the "prior attraction" expected under many efficient-coding models. We propose a gain-adaptive, recurrent sensory network model in which gains optimize a novel efficient-coding objective balancing accuracy and spiking cost. From the propagation of modulated gains throughout the network emerge quickly adaptive tuning curves. The model accounts for subtle adapter-repulsion effects under peaked priors and predicts fast prior attraction under broader distributions, for which we provide supporting behavioral evidence. Our framework reconciles seemingly contradictory adaptive phenomena, under a unified theoretical and mechanistic model of efficient coding mediated by gain modulation in recurrent circuits.

neuroscience↗

bioRxiv2025

Neural and behavioral signatures of policy compression in cognitive control

Making context-dependent decisions incurs cognitive costs. Cognitive control studies have investigated the nature of such costs from both computational and neural perspectives. In this paper, we offer an information-theoretic account of the costs associated with context-dependent decisions. According to this account, the brains limited capacity to store context-dependent policies necessitates "compression" of policies into internal representations with an upper bound on code length, quantified by an information-theoretic measure (policy complexity). These representations are decoded into actions by sequentially inspecting each bit, such that longer codes take more time to decode. When a response deadline is imposed, the account predicts that policy complexity should increase with the deadline. Higher policy complexity is associated with several behavioral signatures: (i) higher accuracy; (ii) lower variability; and (iii) lower perseveration. Analyzing data from a rule-based action selection task, we found evidence supporting all of these predictions. We further hypothesized that complex policies require higher neural dimensionality (which constrains the code space). Consistent with this hypothesis, we found that policy complexity correlates with a measure of neural dimensionality in a rule-based decision task. This finding brings us a step closer to understanding the neural implementation of policy compression and its implications for cognitive control.

neuroscience↗

bioRxiv2025

Reconciling time and prediction error theories of associative learning

Learning involves forming associations between sensory events that have a consistent temporal relationship. Influential theories based on prediction errors explain numerous behavioral and neurobiological observations but do not account for how animals measure the passage of time. Here, we propose a Bayesian theory for temporal causal learning, where the structure of inter-stimulus intervals is used to infer the singular cause of a rewarding stimulus. We show that a single assumption of timescale invariance, formulated as an hierarchical generative model, is sufficient to explain a puzzling set of learning phenomena, including the power-law dependence of acquisition on inter-trial intervals and timescale invariance in response profiles. A biologically plausible algorithm for inference recapitulates salient aspects of both timing and prediction error theories. The theory predicts neural signals with distinct dynamics that encode causal associations and temporal structure.

animal behavior and cognition↗

bioRxiv2024

Policy complexity suppresses dopamine responses

Limits on information processing capacity impose limits on task performance. We show that animals achieve performance on a perceptual decision task that is near-optimal given their capacity limits, as measured by policy complexity (the mutual information between states and actions). This behavioral profile could be achieved by reinforcement learning with a penalty on high complexity policies, realized through modulation of dopaminergic learning signals. In support of this hypothesis, we find that policy complexity suppresses midbrain dopamine responses to reward outcomes, thereby reducing behavioral sensitivity to these outcomes. Our results suggest that policy compression shapes basic mechanisms of reinforcement learning in the brain.

neuroscience↗