Associative Learning Mechanisms Research Articles

Mitigation measures for pinniped-fisheries interactions based on knowledge of animal behavior

Abstract A rise in pinniped-fisheries interactions has been observed due to the global decline in fish populations primarily attributed to overfishing, and/or due to the recent population recovery of several pinniped species. Although many studies have quantified these interactions, a limited number delve into their behavioral aspects. We consulted 374 studies on pinniped-fisheries interactions from 1980 to 2020 and analyzed which of them addressed the interaction from a behavioral perspective. Additionally, we examined the role of learning in pinnipeds and their ability to capture prey from fishing gear. The stimulus-reward relationship in operational interactions was addressed in 51 (14%) studies, involving 10 (19%) of the world’s 54 pinniped species or subspecies. Depredation behavior (43%) and attack behavior (35%) were frequently reported during fishing interactions. Understanding associative and non-associative learning mechanisms has the potential to reduce economic losses from fishing activity, advance efficient mitigation measures to minimize conflicts, and establish pragmatic conservation priorities based on a better understanding of the behavioral aspects of these interactions. More research is needed to explore the effectiveness of deterrents in different species, improve methods of reducing habituation, and evaluate long-term behavioral effects, fostering effective strategies for minimizing conflicts and promoting coexistence between humans and pinnipeds.

Distortions in Judged Spatial Relations in Large Language Models

We present a benchmark for assessing the capability of large language models (LLMs) to discern intercardinal directions between geographic locations and apply it to three prominent LLMs: GPT-3.5, GPT-4, and Llama-2. This benchmark specifically evaluates whether LLMs exhibit a hierarchical spatial bias similar to humans, where judgments about individual locations’ spatial relationships are influenced by the perceived relationships of the larger groups that contain them. To investigate this, we formulated fourteen questions focusing on well-known U.S. cities. Seven questions were designed to challenge the LLMs with scenarios potentially influenced by the orientation of larger geographical units, such as states or countries, whereas the remaining seven targeted locations were less susceptible to such hierarchical categorization. Among the tested models, GPT-4 exhibited superior performance with 55 percent accuracy, followed by GPT-3.5 at 47 percent and Llama-2 at 45 percent. The models showed significantly reduced accuracy on tasks with suspected hierarchical bias. For example, GPT-4’s accuracy dropped to 33 percent on these tasks, compared to 86 percent on others. The models identified the nearest cardinal direction in most cases, however, reflecting their associative learning mechanism, thereby embodying human-like misconceptions. We discuss avenues for improving the spatial reasoning capabilities of LLMs.

Decomposing geographical judgments into spatial, temporal and linguistic components

When mentally exploring maps representing large-scale environments (e.g., countries or continents), humans are assumed to mainly rely on spatial information derived from direct perceptual experience (e.g., prior visual experience with the geographical map itself). In the present study, we rather tested whether also temporal and linguistic information could account for the way humans explore and ultimately represent this type of maps. We quantified temporal distance as the minimum time needed to travel by train across Italian cities, while linguistic distance was retrieved from natural language through cognitively plausible AI models based on non-spatial associative learning mechanisms (i.e., distributional semantic models). In a first experiment, we show that temporal and linguistic distances capture with high-confidence real geographical distances. Next, in a second behavioral experiment, we show that linguistic information can account for human performance over and above real spatial information (which plays the major role in explaining participants’ performance) in a task in which participants have to judge the distance between cities (while temporal information was found to be not relevant). These findings indicate that, when exploring maps representing large-scale environments, humans do take advantage of both perceptual and linguistic information, suggesting in turn that the formation of cognitive maps possibly relies on a strict interplay between spatial and non-spatial learning principles.

Don't throw the associative baby out with the Bayesian bathwater: Children are more associative when reasoning retrospectively under information processing demands.

Causal reasoning is a fundamental cognitive ability that enables individualsto learn about the complex interactions in the world around them. However, the mechanisms that underpin causal reasoning are not well understood. For example, it remains unresolved whether children's causal inferences are best explained by Bayesian inference or associative learning. The two experiments and computational models reported here were designed to examine whether 5- and 6-year-olds will retrospectively reevaluate objects-that is, adjust their beliefs about the causal status of some objects presented at an earlier point in time based on the observed causal status of other objects presented at a later point in time-when asked to reason about 3 and 4 objects and under varying degrees of information processing demands. Additionally, the experiments and models were designed to determine whether children's retrospective reevaluations were best explained by associative learning, Bayesian inference, or some combination of both. The results indicated that participants retrospectively reevaluated causal inferences under minimal information-processing demands (Experiment 1) but failed to do so under greater information processing demands (Experiment 2) and that their performance was better captured by an associative learning mechanism, with less support for descriptions that rely on Bayesian inference. RESEARCH HIGHLIGHTS: Five- and 6-year-old children engage in retrospective reevaluation under minimal information-processing demands (Experiment 1). Five- and 6-year-old children do not engage in retrospective reevaluation under more extensive information-processing demands (Experiment 2). Across both experiments, children's retrospective reevaluations were better explained by a simple associative learning model, with only minimal support for a simple Bayesian model. These data contribute to our understanding of the cognitive mechanisms by which children make causal judgements.

Synaptic mechanisms for associative learning in the cerebellar nuclei

Associative learning during delay eyeblink conditioning (EBC) depends on an intact cerebellum. However, the relative contribution of changes in the cerebellar nuclei to learning remains a subject of ongoing debate. In particular, little is known about the changes in synaptic inputs to cerebellar nuclei neurons that take place during EBC and how they shape the membrane potential of these neurons. Here, we probed the ability of these inputs to support associative learning in mice, and investigated structural and cell-physiological changes within the cerebellar nuclei during learning. We find that optogenetic stimulation of mossy fiber afferents to the anterior interposed nucleus (AIP) can substitute for a conditioned stimulus and is sufficient to elicit conditioned responses (CRs) that are adaptively well-timed. Further, EBC induces structural changes in mossy fiber and inhibitory inputs, but not in climbing fiber inputs, and it leads to changes in subthreshold processing of AIP neurons that correlate with conditioned eyelid movements. The changes in synaptic and spiking activity that precede the CRs allow for a decoder to distinguish trials with a CR. Our data reveal how structural and physiological modifications of synaptic inputs to cerebellar nuclei neurons can facilitate learning.

Dopamine-Dependent Plasticity Is Heterogeneously Expressed by Presynaptic Calcium Activity across Individual Boutons of the Drosophila Mushroom Body.

The Drosophila mushroom body (MB) is an important model system for studying the synaptic mechanisms of associative learning. In this system, coincidence of odor-evoked calcium influx and dopaminergic input in the presynaptic terminals of Kenyon cells (KCs), the principal neurons of the MB, triggers long-term depression (LTD), which plays a critical role in olfactory learning. However, it is controversial whether such synaptic plasticity is accompanied by a corresponding decrease in odor-evoked calcium activity in the KC presynaptic terminals. Here, we address this question by inducing LTD by pairing odor presentation with optogenetic activation of dopaminergic neurons (DANs). This allows us to rigorously compare the changes at the presynaptic and postsynaptic sites in the same conditions. By imaging presynaptic acetylcholine release in the condition where LTD is reliably observed in the postsynaptic calcium signals, we show that neurotransmitter release from KCs is depressed selectively in the MB compartments innervated by activated DANs, demonstrating the presynaptic nature of LTD. However, total odor-evoked calcium activity of the KC axon bundles does not show concurrent depression. We further conduct calcium imaging in individual presynaptic boutons and uncover the highly heterogeneous nature of calcium plasticity. Namely, only a subset of boutons, which are strongly activated by associated odors, undergo calcium activity depression, while weakly responding boutons show potentiation. Thus, our results suggest an unexpected nonlinear relationship between presynaptic calcium influx and the results of plasticity, challenging the simple view of cooperative actions of presynaptic calcium and dopaminergic input.

The pigeon as a machine: Complex category structures can be acquired by a simple associative model

Never known for its smarts, the pigeon has proven to be a prodigious classifier of complex visual stimuli. What explains its surprising success? Does it possess elaborate executive functions akin to those deployed by humans? Or does it effectively deploy an unheralded, but powerful associative learning mechanism? In a series of experiments, we first confirm that pigeons can learn a variety of category structures- some devised to foil the use of advanced cognitive processes. We then contrive a simple associative learning model to see how effectively the model learns the same tasks given to pigeons. The close fit of the associative model to pigeons' categorization behavior provides unprecedented support for associative learning as a viable mechanism for mastering complex category structures and for the pigeon's using this mechanism to adapt to a rich visual world. This model will help guide future neuroscientific research into the biological substrates of visual cognition.

Stress, associative learning, and decision-making

Exposure to acute and chronic stress has significant effects on the basic mechanisms of associative learning and memory. Stress can both impair and enhance associative learning depending on type, intensity, and persistence of the stressor, the subject’s sex, the context that the stress and behavior is experienced in, and the type of associative learning taking place. In some cases, stress can cause or exacerbate the maladaptive behavior that underlies numerous psychiatric conditions including anxiety disorders, obsessive–compulsive disorder, post-traumatic stress disorder, substance use disorder, and others. Therefore, it is critical to understand how the varied effects of stress, which may normally facilitate adaptive behavior, can also become maladaptive and even harmful. In this review, we highlight several findings of associative learning and decision-making processes that are affected by stress in both human and non-human subjects and how they are related to one another. An emerging theme from this work is that stress biases behavior towards less flexible strategies that may reflect a cautious insensitivity to changing contingencies. We consider how this inflexibility has been observed in different associative learning procedures and suggest that a goal for the field should be to clarify how factors such as sex and previous experience influence this inflexibility.

Effects of Experiencing CS-US Pairings on Instructed Fear Reversal.

Fear learning allows us to identify and anticipate aversive events and adapt our behavior accordingly. This is often thought to rely on associative learning mechanisms where an initially neutral conditioned stimulus (CS) is repeatedly paired with an aversive unconditioned stimulus (US), eventually leading to the CS also being perceived as aversive and threatening. Importantly, however, humans also show verbal fear learning. Namely, they have the ability to change their responses to stimuli rapidly through verbal instructions about CS-US pairings. Past research on the link between experience-based and verbal fear learning indicated that verbal instructions about a reversal of CS-US pairings can fully override the effects of previously experienced CS-US pairings, as measured through fear ratings, skin conductance, and fear-potentiated startle. However, it remains an open question whether such instructions can also annul learned CS representations in the brain. Here, we used a fear reversal paradigm (female and male participants) in conjunction with representational similarity analysis of fMRI data to test whether verbal instructions fully override the effects of experienced CS-US pairings in fear-related brain regions or not. Previous research suggests that only the right amygdala should show lingering representations of previously experienced threat ("pavlovian trace"). Unexpectedly, we found evidence for the residual effect of prior CS-US experience to be much more widespread than anticipated, in the amygdala but also cortical regions like the dorsal anterior cingulate or dorsolateral prefrontal cortex. This finding shines a new light on the interaction of different fear learning mechanisms, at times with unexpected consequences.SIGNIFICANCE STATEMENT Humans are able to learn about aversive stimuli both from experience (i.e., repeated pairings of conditioned stimulus (CS) and unconditioned stimulus (US; pavlovian conditioning), and from verbal instructions about stimulus pairings. Understanding how experience-based and verbal learning processes interact is key for understanding the cognitive and neural underpinnings of fear learning. We tested whether prior aversive experiences (CS-US pairings) affected subsequent verbal learning, searching for lingering threat signals after verbal instructions reversed a CS from being threatening to being safe. While past research suggested such threat signals can only be found in the amygdala, we found evidence to be much more widespread, including the medial and lateral PFC. This highlights how experience-based and verbal learning processes interact to support adaptive behavior.

Simple questions on simple associations: regularity extraction in non-human primates

When human and non-human animals learn sequences, they manage to implicitly extract statistical regularities through associative learning mechanisms. In two experiments conducted with a non-human primate species (Guinea baboons, Papio papio), we addressed simple questions on the learning of simple AB associations appearing in longer noisy sequences. Using a serial reaction time task, we manipulated the position of AB within the sequence, such that it could be either fixed (by appearing always at the beginning, middle, or end of a four-element sequence; Experiment 1) or variable (Experiment 2). We also tested the effect of sequence length in Experiment 2 by comparing the performance on AB when it was presented at a variable position within a sequence of four or five elements. The slope of RTs from A to B was taken for each condition as a measurement of learning rate. While all conditions differed significantly from a no-regularity baseline, we found strong evidence that the learning rate did not differ between the conditions. These results indicate that regularity extraction is not impacted by the position of the regularity within a sequence and by the length of the sequence. These data provide novel general empirical constraints for modeling associative mechanisms in sequence learning.

A Computational Model Reveals Learning Dynamics During Interpretation Bias Training With Clinical Applications

BackgroundSome psychopathologies, including anxiety and irritability, are associated with biases when judging ambiguous social stimuli. Interventions targeting these biases, or interpretation bias training (IBT), are amenable to computational modeling to describe their associative learning mechanisms. Here, we translated ALCOVE (attention learning covering map), a model of category learning, to describe learning in youths with affective psychopathology when training on more positive judgments of ambiguous face emotions. MethodsA predominantly clinical sample comprised 71 youths (age range, 8–22 years) representing broad distributions of irritability and anxiety symptoms. Of these, 63 youths were included in the test sample by completing an IBT task with acceptable performance for computational modeling. We used a separate sample of 28 youths to translate ALCOVE for individual estimates of learning rate and generalization. In the test sample, we assessed associations between model learning estimates and irritability, anxiety, their shared variance (negative affectivity), and age. ResultsAge and affective symptoms were associated with category learning during IBT. Lower learning rates were associated with higher negative affectivity common in anxiety and irritability. Lower generalization, or improved discrimination between face emotions, was associated with increasing age. ConclusionsThis work demonstrates a functional consequence of age- and symptom-related learning during interpretation bias. Learning measured by ALCOVE also revealed learning types not accounted for in the prior literature on IBT. This work more broadly demonstrates the utility of measurement models for understanding trial-by-trial processes and identifying individual learning styles.

Associative learning mechanism for drug‐target interaction prediction

AbstractAs a necessary process of modern drug development, finding a drug compound that can selectively bind to a specific protein is highly challenging and costly. Exploring drug‐target interaction strength in terms of drug‐target affinity (DTA) is an emerging and effective research approach for drug development. However, it is challenging to model drug‐target interactions in a deep learning manner, and few studies provide interpretable analysis of models. This paper proposes a DTA prediction method (mutual transformer‐drug target affinity [MT‐DTA]) with interactive learning and an autoencoder mechanism. The proposed MT‐DTA builds a variational autoencoders system with a cascade structure of the attention model and convolutional neural networks. It not only enhances the ability to capture the characteristic information of a single molecular sequence but also establishes the characteristic expression relationship for each substructure in a single molecular sequence. On this basis, a molecular information interaction module is constructed, which adds information interaction paths between molecular sequence pairs and complements the expression of correlations between molecular substructures. The performance of the proposed model was verified on two public benchmark datasets, KIBA and Davis, and the results confirm that the proposed model structure is effective in predicting DTA. Additionally, attention transformer models with different configurations can improve the feature expression of drug/protein molecules. The model performs better in correctly predicting interaction strengths compared with state‐of‐the‐art baselines. In addition, the diversity of drug/protein molecules can be better expressed than existing methods such as SeqGAN and Co‐VAE to generate more effective new drugs. The DTA value prediction module fuses the drug‐target pair interaction information to output the predicted value of DTA. Additionally, this paper theoretically proves that the proposed method maximises evidence lower bound for the joint distribution of the DTA prediction model, which enhances the consistency of the probability distribution between actual and predicted values. The source code of proposed method is available at https://github.com/Lamouryz/Code/tree/main/MT‐DTA.

Resolving the associative learning paradox by category learning in pigeons

A wealth of evidence indicates that humans can engage two types of mechanisms to solve category-learning tasks: declarative mechanisms, which involve forming and testing verbalizable decision rules, and associative mechanisms, which involve gradually linking stimuli to appropriate behavioral responses.1,2,3 In contrast to declarative mechanisms, associative mechanisms have received surprisingly little attention in the broader category-learning literature. Although various forms of associatively driven artificial intelligence (AI) have matched-and even surpassed-humans' performance on several challenging problems,3,4,5,6 associative learning is routinely dismissed as being too simple to power the impressive cognitive achievements of both humans and non-human species.6,7,8,9 Here, we attempt to resolve this paradox by demonstrating that pigeons-which appear to rely solely on associative learning mechanisms in several tasks that promote declarative rule use by humans3,10,11,12-succeed at learning a novel, highly demanding category structure that ought to hinder declarative rule use: the sectioned-rings task. Our findings highlight the power and flexibility that associative mechanisms afford in the realm of category learning.

Associative learning accounts for recursive-structure generation in crows.

Recursive sequence generation (i.e., the ability to transfer recursive patterns to novel items) was recently reported in crows (Liao et al., 2022, Science Advances, 8[44], eabq3356). Here, we argue that although the reported data are certainly compatible with the recursion hypothesis, they can also be explained by other, much simpler mechanisms of associative learning.

Gradient-Based Neuromorphic Learning on Dynamical RRAM Arrays

We present MEMprop, the adoption of gradient-based learning to train fully memristive spiking neural networks (MSNNs). Our approach harnesses intrinsic device dynamics to trigger naturally arising voltage spikes. These spikes emitted by memristive dynamics are analog in nature, and thus fully differentiable, which eliminates the need for surrogate gradient methods that are prevalent in the spiking neural network (SNN) literature. Memristive neural networks typically either integrate memristors as synapses that map offline-trained networks, or otherwise rely on associative learning mechanisms to train networks of memristive neurons. We instead apply the backpropagation through time (BPTT) training algorithm directly on analog SPICE models of memristive neurons and synapses. Our implementation is fully memristive, in that synaptic weights and spiking neurons are both integrated on resistive RAM (RRAM) arrays without the need for additional circuits to implement spiking dynamics, e.g., analog-to-digital converters (ADCs) or thresholded comparators. As a result, higher-order electrophysical effects are fully exploited to use the state-driven dynamics of memristive neurons at run time. By moving towards non-approximate gradient-based learning, we obtain highly competitive accuracy amongst previously reported lightweight dense fully MSNNs on several benchmarks.

Input-timing-dependent plasticity at incoming synapses of the mushroom body facilitates olfactory learning in Drosophila

Aversive olfactory conditioning in Drosophila is a valuable model for elucidating the mechanism of associative learning. Much effort has centered around the role of neuroplasticity at the mushroom body (MB)-mushroom body output neuron (MBON) synapses in mapping odors to specific behaviors. By electrophysiological recordings from MB neurons, we discovered a form of input-timing-dependent plasticity at the incoming synapses from projection neurons that controls the efficacy of aversive olfactory memory formation. Importantly, this plasticity is facilitated by the neural activity of PPL1, the neuronal cluster that also modulates MB-MBON connections at the output stage of MB. Unlike the MB-MBON synapses that probably utilize dopamine D1-like receptors, this neuroplasticity is dependent on D2-like receptors that are expressed mainly by γ Kenyon cells noticeably in their somato-dendritic region. The D2-like receptors recruit voltage-gated calcium channels, leading to calcium influx in the soma and dendrites of γ neurons. Together, our results reveal a previously unrecognized synaptic component of the MB circuit architecture that not only could increase the salience of a conditioning odor but also couples the process of memory encoding and valency mapping to drive-associative learning.

Second verse, same as the first: learning generalizable relational concepts through functional repetition.

The ability to learn and flexibly apply sophisticated concepts is thought by many to be what differentiates humans from all other animals. A basic assumption underlying this belief is that some "lower-order" associative learning mechanisms link perceptual events to specific reactions, whereas the kinds of verbalizable concepts that humans form depend on "higher-order" cognitive processes that rely less on perception and more on rational thought. Evidence in support of this interpretation comes largely from experiments in which animals either fail to learn or generalize concepts that humans readily learn, or learn them with great difficulty. Here, we argue that the formation of generalizable relational concepts may depend more on an individual's capacity to shift attention than on the possession of representational processes that are unique to humans. Studies of relational concept learning in non-human animals show that they can learn generalizable concepts when conditions are favorable. In particular, repetition of similar training experiences appears to facilitate attentional redirection, thereby enabling animals to flexibly reenact past events and to judge the similarity of items within stimulus sets. The conditions that promote concept learning in humans may differ substantially from those experienced by most other animals. This does not imply, however, that either (1) conceptual learning mechanisms differ qualitatively from other learning mechanisms, or (2) that the processes that lead to concept formation in humans differ significantly from those present in other species.

Orthographic priming from unrelated primes: Heterogeneous feedforward inhibition predicted by associative learning

A common assumption among models of orthographic processing is that letter-word inhibitory relationships all share the same strength: activity in the letter B has the same impact on a word like RACE as does equivalent activity in the letter F. However, basic associative learning mechanisms imply that the existence of the neighbor word FACE gives more opportunity to learn a negative weight from the neighbor letter F than from the non-neighbor letter B, leading to stronger negative letter-word weights for neighbor than non-neighbor letters. In masked primed lexical decision, therefore, fity, a neighborly prime formed using neighbor letters, should be a more inhibitory prime for RACE than bund (vice versa for LARK). We present simulations of weight learning using Rescorla and Wagner’s (1972) equations and three experiments consistent with this prediction. Further simulations show heterogeneous feedforward connections from letters to words could contribute to phenomena previously attributed to lexical competition.

Nonverbal Action Interpretation Guides Novel Word Disambiguation in 12-Month-Olds.

Whether young infants can exploit sociopragmatic information to interpret new words is a matter of debate. Based on findings and theories from the action interpretation literature, we hypothesized that 12-month-olds should distinguish communicative object-directed actions expressing reference from instrumental object-directed actions indicative of one's goals, and selectively use the former to identify referents of novel linguistic expressions. This hypothesis was tested across four eye-tracking experiments. Infants watched pairs of unfamiliar objects, one of which was first targeted by either a communicative action (e.g., pointing) or an instrumental action (e.g., grasping) and then labeled with a novel word. As predicted, infants fast-mapped the novel words onto the targeted objects after pointing (Experiments 1 and 4) but not after grasping (Experiment 2) unless the grasping action was preceded by an ostensive signal (Experiment 3). Moreover, whenever infants mapped a novel word onto the object indicated by a communicative action, they tended to map a different novel word onto the distractor object, displaying a mutual exclusivity effect. This reliance on nonverbal action interpretation in the disambiguation of novel words indicates that sociopragmatic inferences about reference likely supplement associative and statistical learning mechanisms from the outset of word learning.

Startle Latency as a Potential Marker for Amygdala-Mediated Hyperarousal

BackgroundFear-related disorders are characterized by hyperexcitability in reflexive circuits and maladaptive associative learning mechanisms. The startle reflex is suited to investigate both processes, either by probing it under baseline conditions or by deriving it in fear conditioning studies. In anxiety research, the amplitude of the fear-potentiated startle has been shown to be influenced by amygdalar circuits and has typically been the readout of interest. In schizophrenia research, prolonged startle peak latency under neutral conditions is an established readout, thought to reflect impaired processing speed. We therefore explored whether startle latency is an informative readout for human anxiety research. MethodsWe investigated potential similarities and differences of startle peak latency and amplitude derived from a classical fear conditioning task in a sample of 206 participants with varying severity levels of anxiety disorders and healthy control subjects. We first reduced startle response to stable components and regressed individual amygdala gray matter volumes onto the resulting startle measures. We then probed time, stimulus, and group effects of startle latency. ResultsWe showed that startle latency and startle amplitude were 2 largely uncorrelated measures; startle latency, but not amplitude, showed a sex-specific association with gray matter volume of the amygdala; startle latencies showed a fear-dependent task modulation; and patients with fear-related disorders displayed shorter startle latencies throughout the fear learning task. ConclusionsThese data provide support for the notion that probing startle latencies under threat may engage amygdala-modulated threat processing, making them a complementary marker for human anxiety research.