Emergent neural dynamics and geometry for generalization in a transitive inference task.

Relational cognition-the ability to infer relationships that generalize to novel combinations of objects-is fundamental to human and animal intelligence. Despite this importance, it remains unclear how relational cognition is implemented in the brain due in part to a lack of hypotheses and predictions at the levels of collective neural activity and behavior. Here we discovered, analyzed, and experimentally tested neural networks (NNs) that perform transitive inference (TI), a classic relational task (if A > B and B > C, then A > C). We found NNs that (i) generalized perfectly, despite lacking overt transitive structure prior to training, (ii) generalized when the task required working memory (WM), a capacity thought to be essential to inference in the brain, (iii) emergently expressed behaviors long observed in living subjects, in addition to a novel order-dependent behavior, and (iv) expressed different task solutions yielding alternative behavioral and neural predictions. Further, in a large-scale experiment, we found that human subjects performing WM-based TI showed behavior inconsistent with a class of NNs that characteristically expressed an intuitive task solution. These findings provide neural insights into a classical relational ability, with wider implications for how the brain realizes relational cognition.

Influence of Rule- and Reward-based Strategies on Inferences of Serial Order by Monkeys.

Knowledge of transitive relationships between items can contribute to learning the order of a set of stimuli from pairwise comparisons. However, cognitive mechanisms of transitive inferences based on rank order remain unclear, as are relative contributions of reward associations and rule-based inference. To explore these issues, we created a conflict between rule- and reward-based learning during a serial ordering task. Rhesus macaques learned two lists, each containing five stimuli that were trained exclusively with adjacent pairs. Selection of the higher-ranked item resulted in rewards. "Small reward" lists yielded two drops of fluid reward, whereas "large reward" lists yielded five drops. Following training of adjacent pairs, monkeys were tested on novels pairs. One item was selected from each list, such that a ranking rule could conflict with preferences for large rewards. Differences between the corresponding reward magnitudes had a strong influence on accuracy, but we also observed a symbolic distance effect. That provided evidence of a rule-based influence on decisions. RT comparisons suggested a conflict between rule- and reward-based processes. We conclude that performance reflects the contributions of two strategies and that a model-based strategy is employed in the face of a strong countervailing reward incentive.

Positional inference in rhesus macaques.

Understanding how organisms make transitive inferences is critical to understanding their general ability to learn serial relationships. In this context, transitive inference (TI) can be understood as a specific heuristic that applies broadly to many different serial learning tasks, which have been the focus of hundreds of studies involving dozens of species. In the present study, monkeys learned the order of 7-item lists of photographic stimuli by trial and error, and were then tested on "derived" lists. These derived test lists combined stimuli from multiple training lists in ambiguous ways, sometimes changing their order relative to training. We found that subjects displayed strong preferences when presented with novel test pairs, even when those pairs were drawn from different training lists. These preferences were helpful when test pairs had an ordering congruent with their ranks during training, but yielded consistently below-chance performance when pairs had an incongruent order relative to training. This behavior can be explained by the joint contributions of transitive inference and another heuristic that we refer to as "positional inference." Positional inferences play a complementary role to transitive inferences in facilitating choices between novel pairs of stimuli. The theoretical framework that best explains both transitive and positional inferences is a spatial model that represents both the position of each stimulus and its uncertainty. A computational implementation of this framework yields accurate predictions about both correct responses and errors on derived lists.

Combining brain perturbation and neuroimaging in non-human primates.

Brain perturbation studies allow detailed causal inferences of behavioral and neural processes. Because the combination of brain perturbation methods and neural measurement techniques is inherently challenging, research in humans has predominantly focused on non-invasive, indirect brain perturbations, or neurological lesion studies. Non-human primates have been indispensable as a neurobiological system that is highly similar to humans while simultaneously being more experimentally tractable, allowing visualization of the functional and structural impact of systematic brain perturbation. This review considers the state of the art in non-human primate brain perturbation with a focus on approaches that can be combined with neuroimaging. We consider both non-reversible (lesions) and reversible or temporary perturbations such as electrical, pharmacological, optical, optogenetic, chemogenetic, pathway-selective, and ultrasound based interference methods. Method-specific considerations from the research and development community are offered to facilitate research in this field and support further innovations. We conclude by identifying novel avenues for further research and innovation and by highlighting the clinical translational potential of the methods.

Category learning in a transitive inference paradigm.

The implied order of a ranked set of visual images can be learned without reliance on information that explicitly signals their order. Such learning is difficult to explain by associative mechanisms, but can be accounted for by cognitive representations and processes such as transitive inference. Our study sought to determine if those processes also apply to learning categories of images. We asked whether participants can (a) infer that stimulus images belonged to familiar categories, even when the images for each trial were unique, and (b) sort those categories into an ordering that obeys transitivity. Participants received minimal verbal instruction and a single session of training. Despite this, they learned the implied order of lists of fixed stimuli and lists of ordered categories, using trial-unique exemplars. We trained two groups, one for which stimuli were constant throughout training and testing (n = 60), and one for which exemplars of each category were trial-unique (n = 50). Our findings suggest that differing cognitive processes may underpin serial learning when learning about specific stimuli as opposed to stimulus categories.

Inferential Learning of Serial Order of Perceptual Categories by Rhesus Monkeys (Macaca mulatta).

Category learning in animals is typically trained explicitly, in most instances by varying the exemplars of a single category in a matching-to-sample task. Here, we show that male rhesus macaques can learn categories by a transitive inference paradigm in which novel exemplars of five categories were presented throughout training. Instead of requiring decisions about a constant set of repetitively presented stimuli, we studied the macaque's ability to determine the relative order of multiple exemplars of particular stimuli that were rarely repeated. Ordinal decisions generalized both to novel stimuli and, as a consequence, to novel pairings. Thus, we showed that rhesus monkeys could learn to categorize on the basis of implied ordinal position, without prior matching-to-sample training, and that they could then make inferences about category order. Our results challenge the plausibility of association models of category learning and broaden the scope of the transitive inference paradigm.SIGNIFICANCE STATEMENT The cognitive abilities of nonhuman animals are of enduring interest to scientists and the general public because they blur the dividing line between human and nonhuman intelligence. Categorization and sequence learning are highly abstract cognitive abilities each in their own right. This study is the first to provide evidence that visual categories can be ordered serially by macaque monkeys using a behavioral paradigm that provides no explicit feedback about category or serial order. These results strongly challenge accounts of learning based on stimulus-response associations.

Functional evolution of new and expanded attention networks in humans.

Macaques are often used as a model system for invasive investigations of the neural substrates of cognition. However, 25 million years of evolution separate humans and macaques from their last common ancestor, and this has likely substantially impacted the function of the cortical networks underlying cognitive processes, such as attention. We examined the homology of frontoparietal networks underlying attention by comparing functional MRI data from macaques and humans performing the same visual search task. Although there are broad similarities, we found fundamental differences between the species. First, humans have more dorsal attention network areas than macaques, indicating that in the course of evolution the human attention system has expanded compared with macaques. Second, potentially homologous areas in the dorsal attention network have markedly different biases toward representing the contralateral hemifield, indicating that the underlying neural architecture of these areas may differ in the most basic of properties, such as receptive field distribution. Third, despite clear evidence of the temporoparietal junction node of the ventral attention network in humans as elicited by this visual search task, we did not find functional evidence of a temporoparietal junction in macaques. None of these differences were the result of differences in training, experimental power, or anatomical variability between the two species. The results of this study indicate that macaque data should be applied to human models of cognition cautiously, and demonstrate how evolution may shape cortical networks.

Implicit Value Updating Explains Transitive Inference Performance: The Betasort Model.

Transitive inference (the ability to infer that B > D given that B > C and C > D) is a widespread characteristic of serial learning, observed in dozens of species. Despite these robust behavioral effects, reinforcement learning models reliant on reward prediction error or associative strength routinely fail to perform these inferences. We propose an algorithm called betasort, inspired by cognitive processes, which performs transitive inference at low computational cost. This is accomplished by (1) representing stimulus positions along a unit span using beta distributions, (2) treating positive and negative feedback asymmetrically, and (3) updating the position of every stimulus during every trial, whether that stimulus was visible or not. Performance was compared for rhesus macaques, humans, and the betasort algorithm, as well as Q-learning, an established reward-prediction error (RPE) model. Of these, only Q-learning failed to respond above chance during critical test trials. Betasort's success (when compared to RPE models) and its computational efficiency (when compared to full Markov decision process implementations) suggests that the study of reinforcement learning in organisms will be best served by a feature-driven approach to comparing formal models.

Representation of outcome risk and action in the anterior caudate nucleus.

The anterior caudate nucleus is essential for goal-directed behavior because it links outcome information to actions. It is well known that caudate neurons provide a variety of reward-related and action signals. However, it is still unclear how the two signals are integrated. We investigated whether and how outcome risk modulates spatial representation. We recorded neural activity in the anterior caudate nucleus while monkeys made saccades to multiple spatial targets, each associated with either fixed (safe) or variable (risky) amount of reward. We report that individual neurons combined the outcome reward signal with spatial information about the direction of saccades. These signals could be reliably read out from the populations of neurons. Moreover, the prospect of a risky outcome improved the quality of spatial information. These results provide direct evidence that global spatial representation in the caudate is modulated by outcome, which can be important for flexible control of behavior, particularly during learning and habit formation, when outcomes vary.

Performance monitoring in monkey frontal eye field.

The frontal eye fields (FEF) are thought to mediate response selection during oculomotor decision tasks. In addition, many FEF neurons have robust postsaccadic responses, but their role in postchoice evaluative processes (online performance monitoring) is only beginning to become apparent. Here we report error-related neural activity in FEF while monkeys performed a biased speed-categorization task that enticed the animals to make impulsive errors. Twenty-three percent of cells in macaque FEF coded an internally generated error-related signal, and many of the same cells also coded task difficulty. The observed responses are primarily consistent with three related concepts that have been associated with performance monitoring: (1) response conflict; (2) uncertainty; and (3) reward prediction. Overall, our findings suggest a novel role for the FEF as part of the neural network that evaluates the preceding choice to optimize behavior in the future.

Smooth pursuit preparation modulates neuronal responses in visual areas MT and MST.

Primates are able to track small moving visual targets using smooth pursuit eye movements. Target motion for smooth pursuit is signaled by neurons in visual cortical areas MT and MST. In this study, we trained monkeys to either initiate or withhold smooth pursuit in the presence of a moving target to test whether this decision was reflected in the relative strength of "go" and "no-go" processes. We found that the gain of the motor response depended strongly on whether monkeys were instructed to initiate or withhold pursuit, thus demonstrating voluntary control of pursuit initiation. We found that the amplitude of the neuronal response to moving targets in areas MT and MST was also significantly lower on no-go trials (by 2.1 spikes/s on average). The magnitude of the neural response reduction was small compared with the behavioral gain reduction. There were no significant differences in neuronal direction selectivity, spatial selectivity, or response reliability related to pursuit initiation or the absence thereof. Variability in eye speed was negatively correlated with firing rate variability after target motion onset during go trials but not during no-go trials, suggesting that MT and MST activity represents an error signal for a negative feedback controller. We speculate that modulation of the visual motion signals in areas MT and MST may be one of the first visual cortical events in the initiation of smooth pursuit and that the small early response modulation may be amplified to produce an all-or-none motor response by downstream areas.

A mathematical theory of relational generalization in transitive inference.

Humans and animals routinely infer relations between different items or events and generalize these relations to novel combinations of items. This allows them to respond appropriately to radically novel circumstances and is fundamental to advanced cognition. However, how learning systems (including the brain) can implement the necessary inductive biases has been unclear. Here we investigated transitive inference (TI), a classic relational task paradigm in which subjects must learn a relation (A > B and B > C) and generalize it to new combinations of items (A > C). Through mathematical analysis, we found that a broad range of biologically relevant learning models (e.g. gradient flow or ridge regression) perform TI successfully and recapitulate signature behavioral patterns long observed in living subjects. First, we found that models with item-wise additive representations automatically encode transitive relations. Second, for more general representations, a single scalar "conjunctivity factor" determines model behavior on TI and, further, the principle of norm minimization (a standard statistical inductive bias) enables models with fixed, partly conjunctive representations to generalize transitively. Finally, neural networks in the "rich regime," which enables representation learning and has been found to improve generalization, unexpectedly show poor generalization and anomalous behavior. We find that such networks implement a form of norm minimization (over hidden weights) that yields a local encoding mechanism lacking transitivity. Our findings show how minimal statistical learning principles give rise to a classical relational inductive bias (transitivity), explain empirically observed behaviors, and establish a formal approach to understanding the neural basis of relational abstraction.

Alteration of functional connectivity in the cortex and major brain networks of non-human primates following focused ultrasound exposure in the dorsal striatum.

BACKGROUND: Focused ultrasound (FUS) is a non-invasive neuromodulation technology that is being investigated for potential treatment of neurological and psychiatric disorders. FUS combined with microbubbles can temporarily open the intact blood-brain barrier (BBB) of animals and humans, and facilitate drug delivery. FUS exposure, either with or without microbubbles, has been demonstrated to alter the behavior of non-human primates (NHP), and previous studies have demonstrated the transient and long-term effects of FUS neuromodulation on functional connectivity using resting state functional MRI. The behavioral effects of FUS vary depending on whether or not it is applied in conjunction with microbubbles to open the BBB, but it is unknown whether opening the BBB affects functional connectivity differently than FUS alone. OBJECTIVE: To compare the effects of applying FUS alone (FUS neuromodulation) and FUS with microbubbles (FUS-BBB opening) on changes of resting state functional connectivity in NHP. METHODS: We applied 2 min FUS exposure without (neuromodulation) and with microbubbles (BBB opening) in the dorsal striatum of lightly anesthetized non-human primates, and acquired resting state functional MRI 40 min respectively after FUS exposure. The functional connectivity (FC) in the cortex and major brain networks between the two approaches were measured and compared. RESULTS: When applying FUS exposure to the caudate nucleus of NHP, we found that both FUS neuromodulation can activate FC between caudate and insular cortex, while inhibiting the FC between caudate and motor cortex. FUS-BBB opening can activate FC between the caudate and medial prefrontal cortex, and within the frontotemporal network (FTN). We also found both FUS and FUS-BBB opening can significantly activate FC within the default mode network (DMN). CONCLUSION: The results suggest applying FUS to a deep brain structure can alter functional connectivity in the DMN and FTN, and that FUS neuromodulation and FUS-mediated BBB opening can have different effects on patterns of functional connectivity.

Internally generated error signals in monkey frontal eye field during an inferred motion task.

An internal model for predictive saccades in frontal cortex was investigated by recording neurons in monkey frontal eye field (FEF) during an inferred motion task. Monkeys were trained to make saccades to the extrapolated position of a small moving target that was rendered temporarily invisible and whose trajectory was altered. On approximately two-thirds of the trials, monkeys made multiple saccades while the target was invisible. Primary saccades were correlated with extrapolated target position. Secondary saccades significantly reduced residual errors resulting from imperfect accuracy of the first saccade. These observations suggest that the second saccade was corrective. Because there was no visual feedback, corrective saccades could only be driven by an internally generated error signal. Neuronal activity in the frontal eye field was directionally tuned before both primary and secondary saccades. Separate subpopulations of cells encoded either saccade direction or direction error before the second saccade. These results suggest that FEF neurons encode the error after the first saccade, as well as the direction of the second saccade. Hence, FEF appears to contribute to detecting and correcting movement errors based on internally generated signals.

The dorsal medial frontal cortex is sensitive to time on task, not response conflict or error likelihood.

The dorsal medial frontal cortex (dMFC) is highly active during choice behavior. Though many models have been proposed to explain dMFC function, the conflict monitoring model is the most influential. It posits that dMFC is primarily involved in detecting interference between competing responses thus signaling the need for control. It accurately predicts increased neural activity and response time (RT) for incompatible (high-interference) vs. compatible (low-interference) decisions. However, it has been shown that neural activity can increase with time on task, even when no decisions are made. Thus, the greater dMFC activity on incompatible trials may stem from longer RTs rather than response conflict. This study shows that (1) the conflict monitoring model fails to predict the relationship between error likelihood and RT, and (2) the dMFC activity is not sensitive to congruency, error likelihood, or response conflict, but is monotonically related to time on task.

Negative allosteric modulation of metabotropic glutamate receptor 5 attenuates alcohol self-administration in baboons.

Many of the behavioral symptoms that define alcohol use disorder (AUD) are thought to be mediated by amplified glutamatergic activity. As a result, previous preclinical studies have investigated glutamate receptor inhibition as a potential pharmacotherapy for AUD, particularly the metabotropic glutamate receptor 5 (mGlu5). In rodents, mGlu5 negative allosteric modulators (NAMs) have been shown to decrease alcohol self-administration. However, their effect on non-human primates has not previously been explored. To bridge this gap, the effects of mGlu5 NAM pretreatment on sweetened alcohol (8% w/v in diluted KoolAid) self-administration in female baboons were evaluated. Two different mGlu5 NAMs were tested: 1) 3-2((-Methyl-4-thiazolyl) ethynyl) pyridine (MTEP) which was administered at a dose of 2 mg/kg IM; and 2) auglurant (N-(5-fluoropyridin-2-yl)-6-methyl-4-(pyrimidin-5-yloxy)picolinamide), a newly developed NAM, which was tested under two different routes (0.001, 0.01, 0.03, 0.1 mg/kg IM and 0.1, 0.3, 1.0 mg/kg PO). MTEP decreased both fixed ratio and progressive ratio responding for sweetened alcohol. Auglurant, administered IM, decreased alcohol self-administration at doses that did not affect self-administration of an alcohol-free sweet liquid reward (0.01 to 0.1 mg/kg). Oral administration of auglurant was not effective in decreasing alcohol self-administration. Our results extend positive findings from rodent studies on mGlu5 regulation of alcohol drinking to female baboons and further strengthen the rationale for targeting mGlu5 in clinical trials for AUD.

Transitive inference after minimal training in rhesus macaques (Macaca mulatta).

Rhesus macaques, when trained for several hundred trials on adjacent items in an ordered list (e.g., A > B, B > C, C > D), are able to make accurate transitive inferences (TI) about previously untrained pairs (e.g., A > C, B > D). How that learning unfolds during training, however, is not well understood. We sought to measure the relationship between the amount of TI training and the resulting response accuracy in 4 rhesus macaques using seven-item lists. The training conditions included the absolute minimal case of presenting each of the six adjacent pairs only once prior to testing. We also tested transfer to nonadjacent pairs with 24 and 114 training trials. Because performance during and after small amounts of training is expected to be near chance levels, we developed a descriptive statistical model to estimate potentially subtle learning effects in the presence of much larger random response variability and systematic bias. These results suggest that subjects learned serial order in an incremental fashion. Thus, rather than performing transitive inference by a logical process, serial learning in rhesus macaques proceeds in a manner more akin to a statistical inference, with an initial uncertainty about list position that gradually becomes more accurate as evidence accumulates. (PsycInfo Database Record (c) 2021 APA, all rights reserved).

Safety evaluation of a clinical focused ultrasound system for neuronavigation guided blood-brain barrier opening in non-human primates.

An emerging approach with potential in improving the treatment of neurodegenerative diseases and brain tumors is the use of focused ultrasound (FUS) to bypass the blood-brain barrier (BBB) in a non-invasive and localized manner. A large body of pre-clinical work has paved the way for the gradual clinical implementation of FUS-induced BBB opening. Even though the safety profile of FUS treatments in rodents has been extensively studied, the histological and behavioral effects of clinically relevant BBB opening in large animals are relatively understudied. Here, we examine the histological and behavioral safety profile following localized BBB opening in non-human primates (NHPs), using a neuronavigation-guided clinical system prototype. We show that FUS treatment triggers a short-lived immune response within the targeted region without exacerbating the touch accuracy or reaction time in visual-motor cognitive tasks. Our experiments were designed using a multiple-case-study approach, in order to maximize the acquired data and support translation of the FUS system into human studies. Four NHPs underwent a single session of FUS-mediated BBB opening in the prefrontal cortex. Two NHPs were treated bilaterally at different pressures, sacrificed on day 2 and 18 post-FUS, respectively, and their brains were histologically processed. In separate experiments, two NHPs that were earlier trained in a behavioral task were exposed to FUS unilaterally, and their performance was tracked for at least 3 weeks after BBB opening. An increased microglia density around blood vessels was detected on day 2, but was resolved by day 18. We also detected signs of enhanced immature neuron presence within areas that underwent BBB opening, compared to regions with an intact BBB, confirming previous rodent studies. Logistic regression analysis showed that the NHP cognitive performance did not deteriorate following BBB opening. These preliminary results demonstrate that neuronavigation-guided FUS with a single-element transducer is a non-invasive method capable of reversibly opening the BBB, without substantial histological or behavioral impact in an animal model closely resembling humans. Future work should confirm the observations of this multiple-case-study work across animals, species and tasks.

A flashing line can warp your mind.

Keeping pace with a constantly changing world requires the ability to make predictions about the future on a variety of timescales. A very basic example of this is the ability to predict the future location of a moving object in the brief time that it takes to perceive and respond to that object. In this issue of Neuron, experiments by Sundberg, Fallah, and Reynolds reveal a potential neural substrate for making short-range predictions about motion in visual area V4.

A neural representation of categorization uncertainty in the human brain.

The ability to classify visual objects into discrete categories ("friend" versus "foe"; "edible" versus "poisonous") is essential for survival and is a fundamental cognitive function. The cortical substrates that mediate this function, however, have not been identified in humans. To identify brain regions involved in stimulus categorization, we developed a task in which subjects classified stimuli according to a variable categorical boundary. Psychophysical functions were used to define a decision variable, categorization uncertainty, which was systematically manipulated. Using event-related functional MRI, we discovered that activity in a fronto-striatal-thalamic network, consisting of the medial frontal gyrus, anterior insula, ventral striatum, and dorsomedial thalamus, was modulated by categorization uncertainty. We found this network to be distinct from the frontoparietal attention network, consisting of the frontal and parietal eye fields, where activity was not correlated with categorization uncertainty.