Frontiers in Behavioral Neuroscience

Operant paradigms are powerful tools to quantify motivation and reward. Traditionally, operant conditioning research has been limited to food and drug reinforcers. Recent advances in commercially available operant equipment, however, allow for the quantification of social motivation. These operant assays are an improvement over commonly used social preference tasks, as they enable direct measurement of the effort and motivation driving social behavior. Based on a design by Venniro et al. (2020), the MedPC social operant boxes modify the traditional operant box setup for social interactions. The experimental rat can lever-press to raise a door for an interaction with a target rat behind a porous barrier. These social operant boxes have been widely adapted to test social behavior in adult and adolescent rodents and investigate how a range of conditions (e.g. stress, drug taking, etc.) affect social motivation. However, there is a gap in assessing maternal motivation for pups during the postpartum period, despite ample evidence that postpartum social behavior is highly relevant for offspring health outcomes. Here, we detail 3D-printed modifications to the standard Med PC social operant boxes to adapt the social target chamber to safely house neonatal pups. We have also developed testing protocols to assess motivation during the limited postpartum period. These data demonstrate that, with simple modifications to social operant chambers and testing protocols, the field can implement advanced behavioral approaches to directly assess maternal motivation.

Motivational deficit is a prevalent symptom across a wide range of neurodegenerative and neuropsychiatric disorders. Despite its clinical importance, first-line treatments for these disorders fail to effectively treat this symptom domain. In animal models, motivation is typically assessed in the context of extrinsic reward, where reward is delivered for completing an effortful action. However, many motivated behaviours occur in the absence of a tangible reward and are instead driven by intrinsic motivation. Previous work has shown that an extrinsic motivation task (effort for reward (EfR)) and an intrinsic motivation task (effort based forage (EBF) task) show opposing responses to a range of pharmacological manipulations. However, it is not clear whether intrinsic and extrinsic motivation dissociate in the context of endogenous behavioural variation. We therefore investigated whether these tasks were sensitive to behavioural variation across three different strains of mice (C57Bl/6JJRi, 129S2/SvPasOrlRj and BALB/cJRi) and whether strain profiles diverged across tasks. Here, we found that BALB/c mice showed the lowest levels of foraging in the EBF task, indicative of a low intrinsic motivational state but showed the highest levels of high effort responding in the EfR task, indicative of a high extrinsic motivational state. These differences were not driven by an anxiety-related phenotype and were therefore indicative of a motivation phenotype divergence across tasks. This work highlights the importance of moving away from considering motivation on a single axis, as findings can diverge depending on the nature of the motivational process. This has important implications for both phenotypic interpretation and the development of treatments targeting motivational dysfunction.

Despite its strong link to neuropsychiatric conditions, frustration remains critically understudied in humans and animals alike. Therefore, there is an urgent need to develop tools to understand and therapeutically target frustration-related functions. Interestingly, humans and rats respond similarly during frustrative nonreward by increasing barpress durations. We previously validated barpress duration in rat operant tasks as a reliable measure of frustration-related behavior; however, it is wellknown that in addition to duration of responding, emotional states such as frustration alter other aspects of responding such as force of pressing. One-dimensional, static measures such as maximum force could miss rich information contained within operant data. Thus, the objective of this study is to apply machine learning (ML) to force/time profiles to discriminate frustration-related barpresses from non-frustration-related barpresses. Results showed an AUROC for FR1 (i.e., non-frustrated) vs. extinction (frustrated condition) for individual barpresses of 0.65 that improved to 0.84 with a chunk size of 10. The model generalized well to progressive ratio responding, a different kind of frustration procedure. We conclude that force/time profiling does provide utility beyond one dimensional measures of duration or force separately, meaning that we can indeed infer the internal state of frustration from behavior using ML techniques. Importantly, this project will also serve as proof-of-concept for applying ML to predict other internal states from barpress data.

Behavioral test batteries are increasingly used to characterize multiple functional domains in zebrafish, yet the potential impact of test sequence on behavioral outcomes remains poorly defined. Here, we systematically evaluated whether test order influences behavioral responses in a three-assay battery comprising the novel tank test (NTT), mirror-induced aggression (MIA), and social preference (SP) test. Adult zebrafish (Danio rerio) were exposed to all possible permutations of the three assays in a fully counterbalanced design, allowing assessment of order effects across locomotor, anxiety-like, aggression-related, and social behaviors. Test order produced modest and parameter-specific effects, primarily affecting locomotor activity in the NTT and social proximity in the SP assay. Time-course analysis revealed within-test behavioral dynamics, with limited evidence that test order modulates early adaptation or late engagement with the testing environment but does not alter overall temporal response profiles. Sex-dependent effects were assay-specific and most pronounced in the NTT, with no consistent sex differences observed in MIA or SP. To evaluate the global structure of behavioral variation, Principal Component Analysis (PCA) was performed across assays. Despite localized effects of test order, no clear multivariate separation between test sequences was observed, indicating that sequential testing does not produce distinct baseline phenotypes. Together, these findings support the robustness and reproducibility of multidomain behavioral batteries while highlighting the importance of standardized test-order reporting to improve cross-study comparability.

Head-fixed behavior in rodents is a foundational technique in systems neuroscience which enables use of sophisticated imaging techniques in combination with animal behavior. However, accessibility of head-fixed behavior techniques is limited. Animal training consumes a large amount of experimenter labor and commercial setups, when available, are largely inflexible and financially burdensome. Here, we present a low-cost, modular, and open-source hardware and software implementation for head-fixed rodent decision-making tasks. Our design lowers experimenter labor and enables large teams of researchers to participate in animal training with minimal experimenter error using a simple touchscreen GUI and automated training progression. We demonstrate the efficacy of the platform by training a cohort of animals in a two-choice probabilistic rapid-reversal task in which mice continuously update action choices based on recent reward history. The presented design lowers the barrier to entry for laboratories seeking to conduct head-fixed rodent behavior and provides modular solutions for developing custom rigs based on experimental demands.

Traditionally, complex behavioral tasks in mice have relied upon water restriction as an external motivator to increase task engagement. However, citric acid (CA) water, the technique whereby water is given ad libitum but made sour by the addition of CA, has emerged as an alternative to typical water restriction. Evidence suggests CA water can effectively motivate task performance while improving animal welfare in alignment with the 3Rs of animal research and reducing experimenter labor. While promising, the applicability of CA water in mice remains incompletely characterized with higher concentrations only tested in rats and "ramping" schedules, where mice progress to increasingly higher concentrations, indirectly examined. Here, we evaluate four CA concentration/schedule combinations for their effects on mouse health (weight changes, home cage behaviors, fecal counts) and motivation to drink regular water (lick counts, drinking behaviors) in female and male C57BL/6J mice. We find that a schedule ramping from 1% to 2% CA after one week is the easiest for mice to adapt to and sustained 2% CA use maintains robust lick counts for at least 5 weeks. Additionally, CA has been directly characterized for wheel-turning and touchscreen tasks, but not virtual reality (VR) tasks, an increasingly popular class of behavioral experiments. Therefore, we also assess how 2% CA affects motivation and task performance in two VR treadmill tasks (running and stopping task). We find that CA water does not improve task performance above that of mice given regular water, but does limit competing motivations and produce more uniform, reward-motivated behavior.

Advances in camera technology and computer vision techniques have allowed researchers to track animals in 3D in ways which previously were difficult or impossible. Many such 3D tracking tools make use of multiple cameras, but unfamiliarity with the principles and technology involved can make it difficult to employ such techniques. In this protocol, we describe Braid, open-source software for live, multi-camera 3D tracking of insects. Using background-subtraction, Braid performs detection of objects without requiring the use of physical markers affixed to the insect. Braid constructs low-latency 3D position estimates using Kalman filtering and nearest neighbor data association. We document in detail the process of tracking freely flying bees within a flight arena using Braid. This protocol includes instructions on installation, configuration of cameras, setup, calibration, and operation. Within the system described here, we demonstrate that Braid can achieve position estimates accurate to <1 millimeter (within a 0.3 cubic meter volume). These factors make Braid suitable for tracking small, fast-flying animals like insects. Braid's low latency allows live tracking, removing the necessity to collect large video files and making it suitable for integration in closed loop systems such as virtual reality. Code is available at https://github.com/strawlab/strand-braid

Drosophila connectomic datasets provide increasingly comprehensive maps of neuronal morphology and synaptic connectivity, offering an unprecedented opportunity to explore the structural organization of its neural circuits. This calls for designing automated tools to interact with connectomic datasets at scale for efficiently exploring structural features embedded in the vast amount of data. Yet the central challenge remains the understanding of the functional logic of neural circuits. In order to understand how elements of the functional logic may emerge from this structural organization, it is critical to (i) characterize the objects in the natural environment in which brain circuits operate, and (ii) formulate how brain circuits represent and process the defined objects in the natural environment. To develop and demonstrate a methodology for these requirements, we focus on the Drosophila looming-evoked escape pathway. We modeled the trajectory of looming objects that are on a collision course (direct-hits) or pass-by the fly (near-misses): their projected images on the retina can be characterized by the solid angle (angular size) and elevation. We then analyzed the pathway's morphology across the OpticLobe, Hemibrain, and FlyWire connectome datasets. By abstracting their sub-neuronal structure and retinotopic organization, we constructed an executable circuit model that maps each structural element to a processing block. We demonstrate that this model separates direct hits from near misses well before the angular size could tell them apart. To accelerate the connectomic analysis step, we developed a Python toolset with an agentic, code-free workspace interface called NeuroGraphBench (NGB). NGB provides four composable morphology-analysis primitives and an AI agent that composes them to interactively respond to natural-language queries aided by visualization on an interactive 3D canvas. Thus, NGB automates tedious and repetitive tasks to enable faster and scalable connectomic exploration, keeping human reasoning, instead of writing code, at the center of an open-ended research inquiry.

RationalePsychedelic compounds such as psilocybin have attracted growing interest for their potential therapeutic effects in psychiatric disorders, with improvements in cognitive flexibility proposed as a possible mechanism of action. However, the effects of psychedelics on cognitive flexibility remain poorly understood. ObjectiveThis study aimed to examine the acute and post-acute effects of psilocybin (0.1, 0.3, 1 mg/kg) and lysergic acid diethylamide (LSD, (0.02, 0.04, 0.08 mg/kg) on cognitive flexibility in male rats. MethodsThis was tested using two variants of the probabilistic reversal learning task (PRLT): a touchscreen-based operant task and a more ethological foraging-based task. ResultsIn the touchscreen PRLT, acute psilocybin disrupted task engagement, with animals completing fewer trials and showing increased trial initiation latency, although psilocybin also showed a trend toward faster initial rule acquisition. However, psilocybin did not significantly alter the number of rule changes achieved, a canonical measure of cognitive flexibility, or feedback sensitivity. LSD similarly produced limited acute effects, although the highest dose reduced lose-shift probability, suggesting decreased sensitivity to negative feedback under some conditions. Post-acute effects of psilocybin were minimal in both PRLT variants and, where LSD effects were observed these occurred across different doses and timepoints without a consistent pattern. ConclusionsOverall, these findings suggest that serotonergic psychedelics do not robustly enhance reversal learning in these paradigms and that apparent learning effects may reflect transient disruptions in task engagement rather than improvements in cognitive flexibility. These results also highlight potential limitations of these PRLT paradigms for detecting psychedelic-induced changes in cognitive flexibility in rodents.

Dominance hierarchies are a fundamental aspect of social organization, enabling animals to minimize aggression and optimize access to resources. Previous studies have highlighted the energetic and physiological demands of dominant status, as well as the behavioral flexibility required of subordinates to navigate these hierarchies. Despite advancements in automated behavior tracking, limitations persist in tracking fine-scale, real-time interactions within complex social environments. Here, we developed and validated a novel RFID-based system to continuously monitor dominance hierarchies in group-housed male mice over 10 days. This system enabled unbiased behavioral inference across light phases and revealed spatial and temporal patterns of dominance behavior undetectable through traditional live-scored methods. Automated tracking accurately identified alpha individuals and consistently inferred linear hierarchies across cohorts, with greater precision for higher-ranked individuals. Behavioral metrics, such as transition frequencies and proximity to food zones, were consistent with dominance driven activity. Hormonal analyses revealed that higher-ranked mice exhibited increased leptin and peptide YY, consistent with heightened activity and satiety signaling, while lower C-peptide levels reflected greater metabolic demands of dominance. Furthermore, dominance rank was associated with differences in light-dark activity, which were in turn related to circulating hormone profiles. This study demonstrates the utility of automated RFID tracking in capturing dominance hierarchies with temporal and spatial granularity, while revealing links between social rank, metabolic regulation, and activity patterns advancing our understanding of social behavior dynamics.

Reinforcement learning theory formulates distinct decision-making strategies, including reactive model-free and deliberative model-based strategies. This study investigates how mice adjust their reinforcement learning strategies while learning decision-making in dynamic environments. Unlike previous studies that focused on behaviors after extensive training periods, we analyzed changes in learning strategies in the course of training of a two-step decision-making task with probabilistic state transition and fluctuating reward probabilities. Our statistical behavioral analysis showed that the stay-probability following common and rare transitions diverged with training, a signature of strategies that utilize knowledge of task structure. We fit various reinforcement learning strategies to behavioral data and found that structure-informed strategies became increasingly dominant in their behaviors during training. Whereas previous studies emphasized transition from goal-directed to habitual strategies after extensive training, which were often associated with model-based and model-free strategies, respectively, our results newly demonstrate a shift from model-free to structure-informed strategies in early training in mice. Author summaryReinforcement learning theory allows us to examine how we make decisions and what approaches we use to optimize rewards. Most previous research, however, has examined animal behavior only after extensive training. Here we analyzed how mice adjust their reinforcement learning strategies as they are trained in a two-step decision-making task. Initially, mice relied on reactive model-free strategies, but as training progressed, their behavior began to incorporate knowledge of task structure. While previous studies suggested transition from model-based to model-free strategies with extensive training, our study revealed the opposite in the early stage of training.

Animals detect and evaluate signs of danger and safety in their environment to ensure survival, yet the neural mechanisms that distinguish safety learning from other forms of conditioned inhibition, remain poorly understood. Here, we directly compared fear and safety learning in male rats. Fear conditioned rats showed high freezing to the tone and the conditioning context, whereas safety conditioned rats showed significant tone-specific reduction in freezing. This safety cue could also generalize to a novel, previously unassociated threat context leading to suppressed freezing when presented demonstrating that inhibitory actions of safety cues are not tied to its original training environment but can modify fear expression across settings. Fear and safety learning also produced unique patterns of neuronal activation and glutamatergic receptor expression in the medial prefrontal cortex (mPFC), basolateral amygdala (BLA), and central amygdala (CeA), as measured by c-Fos immunohistochemistry and Western blotting. Fear conditioning induced greater Fos expression in the BLA and CeA, as well as elevated amygdalar NMDA receptor (GluN1) levels, whereas safety learning increased amygdalar PSD-95 and AMPA receptor (GluA1) expression. Both safety and fear learning increased mPFC Fos expression without affecting glutamatergic receptors levels. Finally, safety conditioning was associated with lower tone-evoked freezing than fear conditioned rats across early extinction sessions and was accompanied by distinct patterns of prefrontal and amygdala activation across extinction. Together, these findings suggest that safety learning engages neural and behavioral mechanisms distinct from fear learning and extinction, while modifying amygdala-prefrontal circuits towards more rapid fear suppression.

"Revealing the structure of pharmacobehavioral space through motion sequencing" by Wiltschko et al. (2020) has been highly influential in behavioral phenotyping research. In a cohort of nearly 700 mice, the authors demonstrated that Motion Sequencing (MoSeq) could distinguish behavioral effects across a large and diverse set of neuroactive and psychoactive compounds. A central conclusion of the study is that MoSeq syllable features substantially outperform more traditional scalar behavioral features in treatment classification tasks. Although this comparison is not emphasized outside the Results section, the reported advantage corresponds to an increase in classification performance exceeding 50% relative to scalar feature representations. While reproducing parts of the analysis using the publicly available dataset, we found that much of this apparent performance difference can be attributed to differences in preprocessing, classifier selection, and hyperparameter optimization. Under alternative, but comparably standard, analytical choices, the performance gap between scalar features and MoSeq syllables was reduced to approximately 11%. Furthermore, in our reanalysis, the performance advantage of MoSeq syllables became statistically significant primarily in highly dense pharmacobehavioral spaces. These findings do not contradict the utility of MoSeq syllables. Rather, they suggest that the magnitude and generality of their advantage over simpler scalar features may depend strongly on analytical methodology and dataset structure. This distinction is practically relevant, as scalar feature approaches are substantially less computationally demanding and often easier to interpret biologically. Consequently, for laboratories with limited computational resources or for studies focused on specific treatment effects, conventional scalar representations may provide a competitive and more accessible alternative. Our findings highlight the importance of analytical standardization and reproducibility in comparative behavioral representation studies.

Understanding the neurobiological systems that regulate alcohol cue-induced craving is of utmost importance for the development of novel intervention strategies for alcohol use disorders (AUDs). However, although a human experimenter is required to conduct alcohol self-administration studies in the lab, the cues associated with the experimenter are seldom if ever factored into the experimental design. Thus, although we have learned much to date about alcohol cue-induced behavior and neurobiology, and in particular about discrete cues presented many times throughout a single daily alcohol self-administration session, we know relatively little about how responses to alcohol availability cues might predict subsequent alcohol consumption. For the current experiment, mice were exposed daily to auditory cues that preceded 2 hours of alcohol or water access using drinking-in-the-dark (DID) methods. An additional control group experienced cues but were not otherwise manipulated. Importantly, cues were initiated remotely from outside the animal facility, avoiding the experimenter being the first cue predicting ethanol availability. Head direction, location in the home cage, and movement velocity were the primary variables on interest. Surprisingly, during the cue period, there were no significant differences between groups in any of these measures, despite meaningful alterations over days. However, we observed many significant correlations between behaviors and drinking variables. First, we observed significant positive associations between ambulatory velocity during cues and subsequent total alcohol (R2=0.14; p<0.0001) and total water (R2=0.12; p=0.0002) consumption, but only in females. We also observed a significant positive relationship (R2=0.25; p<0.0001) between the amount of time oriented toward the sipper port during the auditory cues and the average rate of subsequent alcohol consumption (i.e. front-loading), but only in females. In males, head direction was found to be positively associated with subsequent total water consumption (R2=-0.21; p<0.0001), but not alcohol (R2=-0.01; p=0.2267). We also observed a significant negative relationship (R2=-0.15; p<0.0001) between proximity to the sipper during the cue period and subsequent total 2-hour alcohol intake in males. Although these associations were modest in strength, they suggest potential sex-specific behavioral predictors of alcohol consumption that are regulated by different neural dynamics.

Long-term opioid therapy is limited by analgesic tolerance and opioid-induced hyperalgesia, but the roles of genetic background, sex, and drug exposure remain unclear. We used 20 inbred strains from the Hybrid Rat Diversity Panel to examine thermal sensitivity, oxycodone analgesia, tolerance, and hyperalgesia-like changes following voluntary intravenous oxycodone or saline self-administration. Rats underwent tail-immersion testing before self-administration (Pre-SA) and after self-administration (Post-SA). Oxycodone analgesia was assessed using the percent maximum possible effect time course and the corresponding area under the curve. Pre-SA thermal sensitivity differed across strains and between sexes, and Pre-SA oxycodone analgesia also differed across strains. Oxycodone self-administration produced a sex-dependent increase in thermal sensitivity that was most evident in males. During Post-SA testing, oxycodone self-administering rats showed reduced analgesic responsiveness compared with saline controls, and the magnitude of this difference varied across strains. Within-strain Pre-SA-to-Post-SA comparisons identified tolerance-like reductions in several strains. Across strains and sexes, oxycodone self-administering rats showed a greater Pre-SA-to-Post-SA reduction in analgesic responsiveness than saline controls, consistent with analgesic tolerance. Total oxycodone intake was not associated with tolerance at either the strain-mean or individual-animal level. Heritability estimates were higher for thermal sensitivity and analgesia (H2 {approx} 0.28-0.40) than for changes in thermal sensitivity and tolerance (H2 {approx} 0.18-0.27). These findings demonstrate strain variation in thermal sensitivity and oxycodone analgesia, sex-dependent hyperalgesia-like effects, and reduced analgesic responsiveness following voluntary oxycodone intake.

Acute brain injury disrupts neuro-glial networks leading to impaired brain function. Although injury induces diverse forms of plasticity, their contributions to brain injury outcome remain poorly understood. We previously showed that targeted stab lesions to the optic lobe (OL) of the adult fly brain induce proliferation of glial and neural progenitor cells. Here, we examined the effect of OL lesions on distinct features of fly behavior, which revealed a specific drop in visual stripe fixation performance acutely after injury, whereas locomotor behavior remained mostly unaffected. Using longitudinal studies of injured individuals, we found that flies significantly regain stripe fixation capacity and idiosyncratic traits one week post injury, suggesting a role for plasticity mechanisms. When the proliferation of adult neural progenitor cells is specifically blocked prior to injury, individuals showed no significant improvements of visual orientation during the identified plasticity window suggesting that progenitor activation may support recovery of stripe approach behavior. Hence the individual tracking of orientation behavior emerges as a suitable quantitative framework for studying functional recovery and interindividual variability in the adult Drosophila brain following brain injury.

Intergroup conflict is one of the most significant issues in human society. In the 1950s, Sherif et al. reported that intergroup conflict could be artificially induced in boys through intergroup competition with tug-of-war and ball games. Since this iconic study, researchers have developed various experimental methods to replicate intergroup competition and/or conflicts. However, although intergroup conflicts in wild animals are often reported, it has been difficult to establish a situation of intergroup conflict in laboratory rodents that is discriminable from aggressive behavior individually. In this study, we established a novel experimental paradigm for intergroup competition in mice in which the members of each group shared objectives and tasks. Adult male ICR/Jcl mice were housed in groups of six, divided into two teams of three and repeatedly performed a competitive Tsunahiki task (tsunahiki means tug-of-war in Japanese). The competitive Tsunahiki task was conducted in an open field divided into two experimental fields, with three ropes stuck to a wall separating the fields. The mice were required to pull two ropes out faster than their opponent team to win, and only the winners could proceed to the reward area separated by a guillotine door. We demonstrated that the experience of the competitive Tsunahiki task induced attack bites selectively toward members of the other team (out-group members). Our findings suggest that intergroup competition induces intergroup conflict in mice, providing a technical breakthrough in elucidating the detailed neuroscientific mechanisms underlying intergroup conflict.

Planning husbandry tasks and experiments with Drosophila melanogaster requires converting a target date into development times that depend on the rearing temperature. This calculation needs to be done for each cross, genotype, and temperature, and the risk of error grows quickly. Available laboratory management tools let users register stocks, crosses, and track them, but they do not create schedules based on a clear, adjustable thermal model. To fill that gap, we developed DrosoTracker, a self-contained web application that works offline and predicts Drosophila development with a thermal summation model recalibrated through regression on data from Powsner (1935) (T0 = 11.78 {degrees}C, DD = 116.38 {degrees}C{middle dot}days, R{superscript 2} = 0.997). The model offers an optional two-level calibration driven by user observations. A wild-type strain first adjusts the model to the laboratorys own conditions. Then each genotype is calibrated against that reference using a random-effects shrinkage estimator that accounts for measurement error and between-batch variability. The model creates schedules for husbandry tasks, evaluates adult cohort survival with the Kaplan-Meier estimator and the log-rank test, and calculates sample size for lifespan studies using Schoenfelds formula. The quantitative components were checked against independent references, including Rs survival package and manual calculations. Ongoing work is focused on validating the calibrated model using cohorts specifically bred for this purpose. DrosoTracker runs entirely in the browser, stores data locally, and is available in English and Spanish.

BACKGROUND: An individual's adaptation to threatening environments under uncertainty is reflected in stress responses. Predictability (the ability to anticipate events) and controllability (the ability to control outcomes) are central to how one adapts, yet their joint influence on aversive learning remains unclear. METHODS: Thirty healthy adults completed a probabilistic aversive learning task in which cue-outcome contingencies varied across levels of predictability and controllability, i.e. whether shock intensity depended on prediction accuracy. Prediction accuracy, reaction time, subjective stress ratings, and skin conductance responses were recorded throughout. Trial-wise learning dynamics were estimated using the Volatile Kalman Filter. RESULTS: Prediction accuracy reduced as environments became less predictable and negatively associated with higher learning rates across predictability levels, with the strongest relationship observed in highly predictable blocks. Skin conductance responses showed that moderately predictable environments elicited responses like those in highly predictable environments when accurate predictions reduced shock intensity, but resembled responses in unpredictable environments when shock intensity was uncontrollable. Model comparison revealed a double dissociation between subjective stress ratings and skin conductance responses. Subjective ratings were best explained by model-derived volatility when prediction accuracy determined shock intensity and by belief uncertainty when it was independent of prediction accuracy, whereas skin conductance responses showed the reverse pattern. Reaction times were best explained by belief uncertainty when predictions influenced shock intensity. Higher anxiety was associated with elevated learning rates in highly and moderately predictable blocks when predictions did not control shock intensity.

Animals must generalize from limited experience, yet behavioral experiments in the laboratory setting rarely assess whether or how rapidly they generalize. This contrasts with machine learning systems, where generalization is considered a fundamental test of learning, and emphasizes performance evaluation with new in-distribution or out-of-distribution examples. Here, we used an olfactory categorization task to investigate rules of generalization versus memorization in mice. We trained mice to discriminate between two target odorants mixed with a variable number (0-13) of background odors. There are 32766 possible mixture stimuli to be classified, yet mice learn to generalize from as few as 8 unique mixtures. This generalization is not due to limited memory capacity: mice successfully learned to group the same set of mixtures when category labels were randomly shuffled. Analysis of individual variability revealed features in learning dynamics during training that predict performance in the generalization phase. A linear supervised learning algorithm could describe the generalization from few exemplars well, whereas nonlinear classifiers were necessary to explain memorization. Our experiments suggest that mice have an inductive bias towards generalization, consistent with a preference for simple rules, and will memorize only when forced to do so.