Learning Rule Research Articles

Abstract Current research on emotional modeling in multi-agent systems remains relatively scarce, with most existing studies focusing on a single emotional state. This limitation hinders the ability of current models to fully capture and describe the complex and dynamic behavioral patterns exhibited by agents in diverse environments. To address this issue, this study proposes a strategy learning rule based on both emotion and personality traits. In this framework, an agent’s decision-making behavior is primarily influenced by its emotional state and personality characteristics. Emotions are generated based on the agent’s received payoff, strategy interactions, and feedback from neighboring agents. A mechanism for emotional decay is introduced to allow emotional states to evolve over time, thereby preventing irrational decisions caused by short-term emotional fluctuations. Additionally, the role of personality traits in strategy adaptation is explored. Experimental results show that agents with high compliance tend to adopt their neighbors’ strategies and form cooperative clusters, while those with high activity levels accelerate the diffusion of cooperative strategies. Comparative analysis with other learning rules further confirms the superiority of the proposed emotion- and personality-based strategy learning rule.

Rules in the mist: Emerging probabilistic rules in uncertain categorization.

Diverse perceptual biases emerge from Hebbian plasticity in a recurrent neural network model.

ABSTRACT The transition to parenthood presents significant challenges, particularly for primiparous working mothers who must balance work and family responsibilities. This study examines their parenting experiences, competencies and early attitudes with a focus on how they simultaneously manage family and work demands. Using an explanatory sequential mixed methods design, the study initially employed a quantitative phase with the Early Parenting Attitudes Questionnaire (EPAQ) and the Parenting Sense of Competence Scale (PSCS), followed by in‐depth interviews. Results indicate that more than half of the participants reported lower parental competence (54.1%) and early parenting attitudes, including affection and attachment (57.4%), early learning attitude (50.8%) and rules and respect (52.5%). Thematic network analysis by Attride‐Stirling identified three global themes: (1) parental competence and attitude, reflecting the knowledge and skills mothers possess; (2) challenges and barriers, highlighting difficulties faced by primiparous working mothers; and (3) aspects of support mechanisms, emphasizing the role of personal, familial and societal support. The study provides insights to inform future evidence‐based practices and policy‐level interventions to address the unique needs of children and mothers, considering the impact of parental and family dynamics on early childhood development.

The neural correlates of nominal classification rule learning and their individual differences.

While acquisition curves in human learning averaged at the group level display smooth, gradual changes in performance, individual learning curves across cognitive domains reveal sudden, discontinuous jumps in performance. Similar thresholding effects are a hallmark of a range of nonlinear systems which can be explored using simple, abstract models. Here, I investigate discontinuous changes in learning performance using Amari-Hopfield networks with Hebbian learning rules which are repeatedly exposed to a single stimulus. Simulations reveal that the attractor basin size for a target stimulus increases in discrete jumps rather than gradual changes with repeated stimulus exposure. The distribution of the size of these positive jumps in basin size is best approximated by a lognormal distribution, suggesting that the distribution is heavy-tailed. Examination of the transition graph structure for networks before and after basin size changes reveals that newly acquired states are often organized into hierarchically branching tree structures, and that the distribution of branch sizes is best approximated by a power law distribution. The findings suggest that even simple nonlinear network models of associative learning exhibit discontinuous changes in performance with repeated learning which mirror behavioral results observed in humans. Future work can investigate similar mechanisms in more biologically detailed network models, potentially offering insight into the network mechanisms of learning with repeated exposure or practice.

Humans and other animals use powerful reinforcement learning (RL) mechanisms that have been discovered by evolution over many generations of trial and error. By contrast, artificial agents typically learn using hand-crafted learning rules. Despite decades of interest, the goal of autonomously discovering powerful RL algorithms has proven elusive7-12. In this work, we show that it is possible for machines to discover a state-of-the-art RL rule that outperforms manually-designed rules. This was achieved by meta-learning from the cumulative experiences of a population of agents across a large number of complex environments. Specifically, our method discovers the RL rule by which the agent's policy and predictions are updated. In our large-scale experiments, the discovered rule surpassed all existing rules on the well-established Atari benchmark and outperformed a number of state-of-the-art RL algorithms on challenging benchmarks that it had not seen during discovery. Our findings suggest that the RL algorithms required for advanced artificial intelligence may soon be automatically discovered from the experiences of agents, rather than manually designed.

Networks of interconnected neurons display diverse patterns of collective activity. Relating this collective activity to the network's connectivity structure is a key goal of computational neuroscience. We approach this question for clustered networks, which can form via biologically realistic learning rules and allow for the reactivation of learned patterns. Previous studies of clustered networks have focused on metastabilty between fixed points, leaving open the question of whether clustered spiking networks can display richer dynamics—and if so, whether these can be predicted from their connectivity. Here we show that in the limits of large population size and fast inhibition, the combinatorial threshold linear network (CTLN) model is a mean-field theory for inhibition-stabilized nonlinear Hawkes networks with clustered connectivity. The CTLN has a large body of “graph rules” relating network structure to dynamics. By applying these, we can predict the dynamic attractors of our clustered spiking networks from the structure of between-cluster connectivity. This allows us to construct networks displaying a diverse array of nonlinear cluster dynamics, including metastable periodic orbits and chaotic attractors. Relaxing the assumption that inhibition is fast, we see that the CTLN model is still able to predict the activity of clustered spiking networks with reasonable inhibitory timescales. For slow enough inhibition, we observe bifurcations between CTLN-like dynamics and global excitatory/inhibitory oscillations.

Artificial intelligence (AI) is transforming cardiac electrophysiology across the entire care pathway, from arrhythmia detection on 12-lead electrocardiograms (ECGs) and wearables to the guidance of catheter ablation procedures, through to outcome prediction and therapeutic personalization. End-to-end deep learning (DL) models have achieved cardiologist-level performance in rhythm classification and prognostic estimation on standard ECGs, with a reported arrhythmia classification accuracy of ≥95% and an atrial fibrillation detection sensitivity/specificity of ≥96%. The application of AI to wearable devices enables population-scale screening and digital triage pathways. In the electrophysiology (EP) laboratory, AI standardizes the interpretation of intracardiac electrograms (EGMs) and supports target selection, and machine learning (ML)-guided strategies have improved ablation outcomes. In patients with cardiac implantable electronic devices (CIEDs), remote monitoring feeds multiparametric models capable of anticipating heart-failure decompensation and arrhythmic risk. This review outlines the principal modeling paradigms of supervised learning (regression models, support vector machines, neural networks, and random forests) and unsupervised learning (clustering, dimensionality reduction, association rule learning) and examines emerging technologies in electrophysiology (digital twins, physics-informed neural networks, DL for imaging, graph neural networks, and on-device AI). However, major challenges remain for clinical translation, including an external validation rate below 30% and workflow integration below 20%, which represent core obstacles to real-world adoption. A joint clinical engineering roadmap is essential to translate prototypes into reliable, bedside tools.

Curriculum effects in multitask learning through the lens of contextual inference.

A flexible copper iodide (CuI) thin‐film transistor integrated with a chitosan gate dielectric is demonstrated in this article, which is fabricated on a polyethylene terephthalate (PET) substrate via a low‐temperature solution process. The device operation is enabled by the electric double layer effect induced by proton migration in chitosan, achieving low‐voltage driving at 2 V. Additionally, the device is tested under various mechanical bending conditions, demonstrating its flexibility and mechanical reliability. A transition from short‐term memory to long‐term memory is realized by modulating the pulse amplitude. Furthermore, the synaptic plasticity is shown to be tunable by pulse parameters (duration, frequency, number), which is quantitatively correlated with the proton diffusion kinetics in chitosan. These behaviors, resembling biological spike‐timing‐dependent plasticity, are systematically analyzed to establish a framework for neuromorphic computing. The results highlight the potential of the CuI/chitosan platform for flexible neuromorphic electronics, offering insights into adaptive learning rules and biohybrid systems.

The striatum plays a central role in action selection and reinforcement learning, integrating cortical inputs with dopaminergic signals encoding reward prediction errors. While dopamine modulates synaptic plasticity underlying value learning, the mechanisms that enable selective reinforcement of behaviorally relevant stimulus-action associations–the structural credit assignment problem–remain poorly understood, especially in environments with multiple competing stimuli and actions. Here, we present a computational model in which acetylcholine (ACh), released by striatal cholinergic interneurons, acts as a channel-specific gating signal that restricts plasticity to brief temporal windows following action execution. The model implements a biologically plausible three-factor learning rule requiring presynaptic activity, postsynaptic depolarization, and phasic dopamine, with plasticity gated by cholinergic pauses that temporally align with behaviorally relevant events. This mechanism ensures that only synapses involved in the selected behavior are eligible for modification. Through systematic evaluation across tasks with distractors and contingency reversals, we show that ACh-gated learning promotes synaptic specificity, suppresses cross-channel interference, and yields increasingly competitive performance relative to Q-learning in complex tasks, reflecting the scalability of the proposed learning mechanism. Moreover, the model reveals distinct roles for striatal pathways: direct pathway (D1) neurons maintain stimulus-specific responses, while indirect pathway (D2) neurons are progressively recruited to suppress outdated associations during policy adaptation. These findings provide a mechanistic account of how coordinated cholinergic and dopaminergic signaling can support scalable and efficient reinforcement learning in the striatum, consistent with experimental observations of pathway-specific plasticity.

Spiking Neural Networks (SNNs) exhibit their optimal information-processing capability at the edge of chaos, but tuning them to this critical regime in reservoir-computing architectures usually relies on costly trial-and-error or plasticity-driven adaptation. This work presents an analytical framework for configuring in the critical regime a SNN-based reservoir with a highly general topology. Specifically, we derive and solve a mean-field equation that governs the evolution of the average membrane potential in leaky integrate-and-fire neurons, and provide an approximation for the critical point. This framework reduces the need for an extensive online fine-tuning, offering a streamlined path to near-optimal network performance from the outset. Through extensive numerical experiments, we validate the theoretical predictions by analyzing the network’s spiking dynamics and quantifying its computational capacity using the information-based Lempel-Ziv-Welch complexity near criticality. Finally, we explore self-organized quasi-criticality by implementing a local homeostatic learning rule for synaptic weights, demonstrating that the network’s dynamics remain close to the theoretical critical point. Beyond AI, our approach and findings also have significant implications for computational neuroscience, providing a principled framework for quantitatively understanding how (neuro)biological networks exploit criticality for efficient information processing. The paper is accompanied by Python code, enabling the reproducibility of the findings.

Inactivation of primate cortex reveals inductive biases in visual learning.

Current machine learning systems consume vastly more energy than biological brains. Neuromorphic systems aim to overcome this difference by mimicking the brain’s information coding via discrete voltage spikes. However, it remains unclear how both artificial and natural networks of spiking neurons can learn energy-efficient information processing strategies. Here we propose Predictive Coding Light (PCL), a recurrent hierarchical spiking neural network for unsupervised representation learning. In contrast to previous predictive coding approaches, PCL does not transmit prediction errors to higher processing stages. Instead it suppresses the most predictable spikes and transmits a compressed representation of the input. Using only biologically plausible spike-timing based learning rules, PCL reproduces a wealth of findings on information processing in visual cortex and permits strong performance in downstream classification tasks. Overall, PCL offers a new approach to predictive coding and its implementation in natural and artificial spiking neural networks.

Advancing the Biological Plausibility and Efficacy of Hebbian Convolutional Neural Networks.

Neural networks recruited for numerical rule learning and application.

Mining and modeling of interactive association rules for mobile learning in flipped classroom based on Apriori algorithm

In this work, a supervised learning rule based on Temporal Single Spike Coding for Effective Transfer Learning (TS4TL) is presented, an efficient approach for training multilayer fully connected Spiking Neural Networks (SNNs) as classifier blocks within a Transfer Learning (TL) framework. A new target assignment method named as "Absolute Target" is proposed, which utilizes a fixed, non-relative target signal specifically designed for single-spike temporal coding. In this approach, the firing time of the correct output neuron is treated as the target spike time, while no spikes are assigned to the other neurons. Unlike existing relative target strategies, this method minimizes computational complexity, reduces training time, and decreases energy consumption by limiting the number of spikes required for classification, all while ensuring a stable and computationally efficient training process. By seamlessly integrating this learning rule into the TL framework, TS4TL effectively leverages pre-trained feature extractors, demonstrating robust performance even with limited labelled data and varying data distributions. The proposed learning rule scales efficiently across both shallow and deep network architectures while maintaining consistent accuracy and reliability. Extensive evaluations on benchmark datasets highlight the strength of this approach, achieving state-of-the-art accuracies, including 98.91% on Eth80, surpassing previous works, and 91.89% on Fashion-MNIST, outperforming all fully connected structures in the literature. Additionally, high accuracies of 98.45% and 97.75% were recorded on the MNIST and Caltech101-Face/Bike datasets, respectively. Furthermore, TS4TL addresses a critical challenge by effectively reducing neuron misfires, ensuring that neurons respond correctly based on first-spike coding, a significant improvement over manually imposed solutions seen in prior works. These contributions collectively highlight the potential of TS4TL as a scalable and high-performance solution for temporal learning in SNNs.

The fuzzy-associative metaheuristic approach addresses the urgent task of developing a marketing decision support system based on a fuzzy trained associative rules expert system, aimed at improving the accuracy and efficiency of consumer preference analysis. The proposed system combines the interpretability of fuzzy logic with data-driven learning via associative rules and parameter identification using an adaptive multi-agent optimization method. To achieve this goal, associative rule learning techniques (Apriori and FP-Growth) were used to extract frequent consumer behavior patterns. A fuzzy expert system was developed, in which the parameters of membership functions are optimized by the Adaptive Vibrating Particle System (AVPS) metaheuristic. Unlike traditional vibrating particle systems, AVPS integrates iteration-dependent control of particle positions, enabling global search in early iterations and local refinement at later stages, thus improving convergence speed and solution precision. The architecture was implemented using Python-based tools (TensorFlow, Keras, Pandas, mlxtend, Scikit-Fuzzy), and validated on the “Consumer Behavior and Shopping Habits” dataset. The fuzzy expert system achieved an accuracy of 0.98, outperforming human experts (0.80), traditional VPS optimization (0.93), and backpropagation-based training (0.90). The system also reduces reliance on manually tuned parameters and increases robustness to data incompleteness and noise. Scientific novelty lies in combining a fuzzy associative rule-learning framework with AVPS-based optimization, offering a scalable and interpretable decision-making mechanism. The developed system contributes to the advancement of intelligent recommendation engines, personalized marketing tools, and decision support systems in consumer-oriented analytics.