Classical Learning Theory Research Articles

A Review of Classical Self-Regulated Learning Theories

Self-regulated learning is not only a specific behavioral strategy in the learning process of learners, but also equally an excellent learning quality of individual learners. This paper will introduce and analyze the theory of classical self-regulated learning, which can effectively help us understand the essence of self-regulated learning more deeply and deepen some knowledge about self-regulated learning. And the essence and scope of self-regulated learning will be elaborated from the information processing theory, social construction theory, social cognitive theory and volitional theory of self-regulated learning.

Understanding Surprising Generalization Phenomena in Deep Learning

Deep learning has exhibited a number of surprising generalization phenomena that are not captured by classical statistical learning theory. This talk will survey some of my work on the theoretical characterizations of several such intriguing phenomena: (1) Implicit regularization: A major mystery in deep learning is that deep neural networks can often generalize well despite their excessive expressive capacity. Towards explaining this mystery, it has been suggested that commonly used gradient-based optimization algorithms enforce certain implicit regularization which effectively constrains the model capacity. (2) Benign overfitting: In certain scenarios, a model can perfectly fit noisily labeled training data, but still archives near-optimal test error at the same time, which is very different from the classical notion of overfitting. (3) Grokking: In certain scenarios, a model initially achieves perfect training accuracy but no generalization (i.e. no better than a random predictor), and upon further training, transitions to almost perfect generalization. Theoretically establishing these properties often involves making appropriate high-dimensional assumptions on the problem as well as a careful analysis of the training dynamics.

Unveiling the synergy of peer feedback and the Metaverse

Undeniably, technology has become an integral part of education across all disciplines. Over the past decades, significant technological advancements, such as the normalization of computers and the popularization of mobile devices, have transformed and revolutionized our learning paradigms. Education has transitioned from the bricks-and-mortar, teacher fronted, and textbook-based teaching to out-of-class, student-centered, and inquiry-based learning. Since 2020, emerging technologies such as the Metaverse and artificial intelligence have shown great potential in expanding learning opportunities, ranging from the location of learning and teaching, the student-teacher machine interactions, to different modes of instruction and learning. In particular, the Metaverse has the power to transform the way peer feedback is provided, which is an essential aspect of effective education. This position paper aims to shed some light on the intersection of peer feedback and the Metaverse. It starts with a review of the classic sociocultural learning theory and the historical use of technology to support peer feedback, followed by an introduction to the definition and features of the Metaverse technology. Then, the paper proposes a tentative model about how the Metaverse can be utilized to facilitate peer feedback. It ends with important questions for educators and learners to contemplate in order to make the most of this new technology in improving the provision and uptake of peer feedback.

Constrained Learning With Non-Convex Losses

Though learning has become a core component of modern information processing, there is now ample evidence that it can lead to biased, unsafe, and prejudiced systems. The need to impose requirements on learning is therefore paramount, especially as it reaches critical applications in social, industrial, and medical domains. However, the non-convexity of most modern statistical problems is only exacerbated by the introduction of constraints. Whereas good unconstrained solutions can often be learned using empirical risk minimization, even obtaining a model that satisfies statistical constraints can be challenging. All the more so, a good one. In this paper, we overcome this issue by learning in the empirical dual domain, where constrained statistical learning problems become unconstrained and deterministic. We analyze the generalization properties of this approach by bounding the empirical duality gap -- i.e., the difference between our approximate, tractable solution and the solution of the original (non-convex) statistical problem -- and provide a practical constrained learning algorithm. These results establish a constrained counterpart to classical learning theory, enabling the explicit use of constraints in learning. We illustrate this theory and algorithm in rate-constrained learning applications arising in fairness and adversarial robustness.

On Generalization Bounds for Deep Networks Based on Loss Surface Implicit Regularization

The classical statistical learning theory implies that fitting too many parameters leads to overfitting and poor performance. That modern deep neural networks generalize well despite a large number of parameters contradicts this finding and constitutes a major unsolved problem towards explaining the success of deep learning. While previous work focuses on the implicit regularization induced by stochastic gradient descent (SGD), we study here how the local geometry of the energy landscape around local minima affects the statistical properties of SGD with Gaussian gradient noise. We argue that under reasonable assumptions, the local geometry forces SGD to stay close to a low dimensional subspace and that this induces another form of implicit regularization and results in tighter bounds on the generalization error for deep neural networks. To derive generalization error bounds for neural networks, we first introduce a notion of stagnation sets around the local minima and impose a local essential convexity property of the population risk. Under these conditions, lower bounds for SGD to remain in these stagnation sets are derived. If stagnation occurs, we derive a bound on the generalization error of deep neural networks involving the spectral norms of the weight matrices but not the number of network parameters. Technically, our proofs are based on controlling the change of parameter values in the SGD iterates and local uniform convergence of the empirical loss functions based on the entropy of suitable neighborhoods around local minima.

On the Hardness of PAC-learning Stabilizer States with Noise

We consider the problem of learning stabilizer states with noise in the Probably Approximately Correct (PAC) framework of Aaronson (2007) for learning quantum states. In the noiseless setting, an algorithm for this problem was recently given by Rocchetto (2018), but the noisy case was left open. Motivated by approaches to noise tolerance from classical learning theory, we introduce the Statistical Query (SQ) model for PAC-learning quantum states, and prove that algorithms in this model are indeed resilient to common forms of noise, including classification and depolarizing noise. We prove an exponential lower bound on learning stabilizer states in the SQ model. Even outside the SQ model, we prove that learning stabilizer states with noise is in general as hard as Learning Parity with Noise (LPN) using classical examples. Our results position the problem of learning stabilizer states as a natural quantum analogue of the classical problem of learning parities: easy in the noiseless setting, but seemingly intractable even with simple forms of noise.

Clustering and enhanced classification using a hybrid quantum autoencoder

Quantum machine learning (QML) is a rapidly growing area of research at the intersection of classical machine learning and quantum information theory. One area of considerable interest is the use of QML to learn information contained within quantum states themselves. In this work, we propose a novel approach in which the extraction of information from quantum states is undertaken in a classical representational-space, obtained through the training of a hybrid quantum autoencoder (HQA). Hence, given a set of pure states, this variational QML algorithm learns to identify—and classically represent—their essential distinguishing characteristics, subsequently giving rise to a new paradigm for clustering and semi-supervised classification. The analysis and employment of the HQA model are presented in the context of amplitude encoded states—which in principle can be extended to arbitrary states for the analysis of structure in non-trivial quantum data sets.

‚Influenc­­­ing‘ – ein Modell religionspädagogisch reflektierten Handelns?

Abstract With the changes in the world we live in, the conditions for learning change. Children and adolescents have access to orientation opportunities in the media, which they can approach independently. Digital environments call for a rethinking of educational theory and teaching practice in terms of content and didactics. In the contexts of YouTube and Instagram, Influencing has developed as a new factor in market, opinion and personality formation. In order to understand peer learning on social media, two classical learning theories can be applied. In the context of communication science and media pedagogy, influencing is based on an update of the conception of parasocial relationship. Technological aspects have to be taken into account. The article concludes with a media pedagogical assessment and religious pedagogical perspectives for educational practice and research.

Binary classification with classical instances and quantum labels

In classical statistical learning theory, one of the most well-studied problems is that of binary classification. The information-theoretic sample complexity of this task is tightly characterized by the Vapnik-Chervonenkis (VC) dimension. A quantum analog of this task, with training data given as a quantum state has also been intensely studied and is now known to have the same sample complexity as its classical counterpart. We propose a novel quantum version of the classical binary classification task by considering maps with classical input and quantum output and corresponding classical-quantum training data. We discuss learning strategies for the agnostic and for the realizable case and study their performance to obtain sample complexity upper bounds. Moreover, we provide sample complexity lower bounds which show that our upper bounds are essentially tight for pure output states. In particular, we see that the sample complexity is the same as in the classical binary classification task w.r.t. its dependence on accuracy, confidence and the VC-dimension.

The Neural Correlates of Cued Reward Omission

Compared to our understanding of positive prediction error signals occurring due to unexpected reward outcomes, less is known about the neural circuitry in humans that drives negative prediction errors during omission of expected rewards. While classical learning theories such as Rescorla–Wagner or temporal difference learning suggest that both types of prediction errors result from a simple subtraction, there has been recent evidence suggesting that different brain regions provide input to dopamine neurons which contributes to specific components of this prediction error computation. Here, we focus on the brain regions responding to negative prediction error signals, which has been well-established in animal studies to involve a distinct pathway through the lateral habenula. We examine the activity of this pathway in humans, using a conditioned inhibition paradigm with high-resolution functional MRI. First, participants learned to associate a sensory stimulus with reward delivery. Then, reward delivery was omitted whenever this stimulus was presented simultaneously with a different sensory stimulus, the conditioned inhibitor (CI). Both reward presentation and the reward-predictive cue activated midbrain dopamine regions, insula and orbitofrontal cortex. While we found significant activity at an uncorrected threshold for the CI in the habenula, consistent with our predictions, it did not survive correction for multiple comparisons and awaits further replication. Additionally, the pallidum and putamen regions of the basal ganglia showed modulations of activity for the inhibitor that did not survive the corrected threshold.

Distribution-Dependent Weighted Union Bound.

In this paper, we deal with the classical Statistical Learning Theory’s problem of bounding, with high probability, the true risk of a hypothesis h chosen from a set of m hypotheses. The Union Bound (UB) allows one to state that where is the empirical errors, if it is possible to prove that and , when h, , and are chosen before seeing the data such that and . If no a priori information is available and are set to , namely equally distributed. This approach gives poor results since, as a matter of fact, a learning procedure targets just particular hypotheses, namely hypotheses with small empirical error, disregarding the others. In this work we set the and in a distribution-dependent way increasing the probability of being chosen to function with small true risk. We will call this proposal Distribution-Dependent Weighted UB (DDWUB) and we will retrieve the sufficient conditions on the choice of and that state that DDWUB outperforms or, in the worst case, degenerates into UB. Furthermore, theoretical and numerical results will show the applicability, the validity, and the potentiality of DDWUB.

Considerations and Suggestions for Design of a Learning and Development Program for Sport Coaches

Coaches play a highly influential role across the whole spectrum of sport. Their proficiency is crucial to the progression, welfare and even the personal development of the athletes under their charge. Maximising the development of that proficiency is therefore a crucial concern of numerous sporting and community organisations. Effort must be directed to continuous improvement of learning and development programs for coaches, with consideration given to the differing requirements of coaches operating in different contexts. Historically, most coach education initiatives have entailed formal instruction in “classroom” situations, but feedback from participants suggests that this is sub-optimal. Better outcomes might be achieved through methods designed to promote informal learning within real-world coaching environments. There is evidence that experiential learning resulting from repeated cycles of action and reflection can contribute substantially to the development of expertise. Observation and imitation of significant others is also a powerful learning mechanism. Here, we review and summarise literature relating to these matters, while also addressing topics such as the nature of coaching, the characteristics of expert coaches, and the ways in which coaches currently learn. We touch upon the concepts of situated learning, cognitive apprenticeship, and mentoring, and identify known barriers and facilitators to adult learning. We then provide a suggested framework for the practical design, implementation, evaluation, and continual refinement of a program emphasising informal but guided nurturing of coach learning and development. The suggested approach is consistent with classical learning theories and incorporates the establishment of critical friendships and communities of practice.

ВЛИЯНИЕ КЛАССИЧЕСКИХ ПОЗНАВАТЕЛЬНЫХ ТЕОРИЙ НА ПРОЦЕСС ОБУЧЕНИЯ

Classical cognitive learning theories are based on the position of the leading and determining role of cognitive processes in learning, which makes it possible to combine various ideas, approaches and research methods in an eclectic way. The analysis of the cognitive approach in educational psychology in the United States allowed us to establish the main directions in it, which are presented in detail in the article. The described cognitive theories can be successfully applied in modern learning conditions, since the development of active thinking becomes a fundamental task of education in the period of modernization and the search for effective ways to optimize the assimilation of a large amount of material and transformation of theoretical knowledge into practical skills.

7. Eco-conectivismo: modelando el conocimiento conectivo

Este artículo presenta una propuesta educativa llamada eco-conectivismo, la cual es una forma de abordar el paradigma pedagógico conectivista en el aula de clases. Iniciamos con una discusión acerca de la influencia de las tecnologías y el aprendizaje informal en la educación formal. A partir de allí, exploramos las limitaciones de las teorías clásicas de aprendizaje para explicar el conocimiento generado en contextos de aprendizaje no estructurados. Abordamos el paradigma conectivista como estrategia para manejar estas limitaciones teóricas, y establecemos las bases de un modelo ontológico que permita su reconocimiento. Asimismo, usamos el paradigma de ecologías de conocimiento, para caracterizar un modelo de analítica de datos que permite desplegar el paradigma conectivista a partir de las interacciones que emergen en un proceso de aprendizaje. Finalmente, explicamos los detalles de su usopara un caso de estudio llamado SaCI, que es un ambiente inteligente de aprendizaje.

Reward Prediction Errors Reflect an Underlying Learning Process That Parallels Behavioural Adaptations: A Trial-to-Trial Analysis

Reinforcement learning can lead to rapid changes in performance. Computational accounts of reinforcement learning align with classic learning theory, as reported by Sutton and Barto (1998, 2018) and suggest that trial-to-trial changes in performance follow rapid but decelerating learning curves. Although there is some support for a link between changes in behavioural and neural data, evidence has been inconclusive. Here, we had a computational model and human participants learn a novel language through trial-and-error while recording electroencephalographic data. By conducting linear mixed-effects models of trial-to-trial analyses, we sought to determine whether neural signals were indicative of a learning process and whether they were related to changes in behaviour. We found that neural measures did diminish with trial-to-trial changes in performance and that they were predictive of behavioural adaptations in both simulated and empirical data. These neural signals are theorised as reward prediction errors—the computational difference between expectations and outcomes—and here we provide compelling evidence that they reflect an underlying learning process that parallels behavioural adaptation.

Climbing Fibers Provide Graded Error Signals in Cerebellar Learning.

The cerebellum plays a critical role in coordinating and learning complex movements. Although its importance has been well recognized, the mechanisms of learning remain hotly debated. According to the classical cerebellar learning theory, depression of parallel fiber synapses instructed by error signals from climbing fibers, drives cerebellar learning. The uniqueness of long-term depression (LTD) in cerebellar learning has been challenged by evidence showing multi-site synaptic plasticity. In Purkinje cells, long-term potentiation (LTP) of parallel fiber synapses is now well established and it can be achieved with or without climbing fiber signals, making the role of climbing fiber input more puzzling. The central question is how individual Purkinje cells extract global errors based on climbing fiber input. Previous data seemed to demonstrate that climbing fibers are inefficient instructors, because they were thought to carry “binary” error signals to individual Purkinje cells, which significantly constrains the efficiency of cerebellar learning in several regards. In recent years, new evidence has challenged the traditional view of “binary” climbing fiber responses, suggesting that climbing fibers can provide graded information to efficiently instruct individual Purkinje cells to learn. Here we review recent experimental and theoretical progress regarding modulated climbing fiber responses in Purkinje cells. Analog error signals are generated by the interaction of varying climbing fibers inputs with simultaneous other synaptic input and with firing states of targeted Purkinje cells. Accordingly, the calcium signals which trigger synaptic plasticity can be graded in both amplitude and spatial range to affect the learning rate and even learning direction. We briefly discuss how these new findings complement the learning theory and help to further our understanding of how the cerebellum works.

反馈负波及其近10年理论解释

<p id="C2">External feedback plays a vital role in ongoing outcome monitoring and future behavioral adjustments. Previous event-related potentials (ERPs) studies have consistently associated the feedback- related negativity (FRN) component with feedback processing after decision making. Until now, several theoretical interpretations of FRN were proposed and continuously updated in the last decade, including the classical reinforcement learning theory, affective-motivational hypothesis, reward positivity (RewP) theory, Predicted response-outcome model (PRO model) and Positive affective model. Whereas different models emphasized on different aspects, no unified theory has yet been proposed to integrate all of the existing experimental evidences. Based on these literature reviews, we argued that big-sample data, multiple and complementary technologies are highly necessary in future FRN studies. Moreover, we proposed that FRN should be considered as an electrophysiological index of reward system to investigate human behaviors in complex social interaction contexts.

The consistency of least-square regularized regression with negative association sequence

In the last few years, many known works in learning theory stepped over the classical assumption that samples are independent and identical distribution and investigated learning performance based on non-independent samples, as mixing sequences (e.g., [Formula: see text]-mixing, [Formula: see text]-mixing, [Formula: see text]-mixing etc.), they derived similar results with the investigation based on classical sample assumption. Negative association (NA) sequence is a kind of significant dependent random variables and plays an important role in non-independent sequences. It is widely applied to various subjects such as probability theory, statistics and stochastic processes. Therefore, it is essential to study the learning performance of learning process for dependent samples drawn from NA process. Obviously, samples in this learning process are not independent and identical distribution. The results in classical learning theory are not applied directly. In this paper, we study the consistency of least-square regularized regression with NA samples. We establish the error bound of least-square regularized regression for NA samples, and prove that the learning rate of least-square regularized regression for NA samples is [Formula: see text], which is tend to [Formula: see text] when [Formula: see text] arbitrarily close to 0, where [Formula: see text] denote the number of the samples. The simulation experiment of convergence rate on NA samples reveals that the least-square regularized regression algorithm for NA samples is consistent. This result generalizes the classical result of independent and identical distribution.

Suppression of Striatal Prediction Errors by the Prefrontal Cortex in Placebo Hypoalgesia

Classical learning theories predict extinction after the discontinuation of reinforcement through prediction errors. However, placebo hypoalgesia, although mediated by associative learning, has been shown to be resistant to extinction. We tested the hypothesis that this is mediated by the suppression of prediction error processing through the prefrontal cortex (PFC). We compared pain modulation through treatment cues (placebo hypoalgesia, treatment context) with pain modulation through stimulus intensity cues (stimulus context) during functional magnetic resonance imaging in 48 male and female healthy volunteers. During acquisition, our data show that expectations are correctly learned and that this is associated with prediction error signals in the ventral striatum (VS) in both contexts. However, in the nonreinforced test phase, pain modulation and expectations of pain relief persisted to a larger degree in the treatment context, indicating that the expectations were not correctly updated in the treatment context. Consistently, we observed significantly stronger neural prediction error signals in the VS in the stimulus context compared with the treatment context. A connectivity analysis revealed negative coupling between the anterior PFC and the VS in the treatment context, suggesting that the PFC can suppress the expression of prediction errors in the VS. Consistent with this, a participant's conceptual views and beliefs about treatments influenced the pain modulation only in the treatment context. Our results indicate that in placebo hypoalgesia contextual treatment information engages prefrontal conceptual processes, which can suppress prediction error processing in the VS and lead to reduced updating of treatment expectancies, resulting in less extinction of placebo hypoalgesia.SIGNIFICANCE STATEMENT In aversive and appetitive reinforcement learning, learned effects show extinction when reinforcement is discontinued. This is thought to be mediated by prediction errors (i.e., the difference between expectations and outcome). Although reinforcement learning has been central in explaining placebo hypoalgesia, placebo hypoalgesic effects show little extinction and persist after the discontinuation of reinforcement. Our results support the idea that conceptual treatment beliefs bias the neural processing of expectations in a treatment context compared with a more stimulus-driven processing of expectations with stimulus intensity cues. We provide evidence that this is associated with the suppression of prediction error processing in the ventral striatum by the prefrontal cortex. This provides a neural basis for persisting effects in reinforcement learning and placebo hypoalgesia.

Social modeling effects on perception of the police

Purpose– The purpose of this paper is to investigate whether indirect police contacts through observational learning models impact students’ trust in the police and their perceptions of police bias.Design/methodology/approach– The study is based on a survey at two public universities in the mid-western and southern regions of the USA (921 out of 1,089 responses were retained for this study). The empirical analysis relied on a principle component factor analysis and a multivariate regression analysis.Findings– Results show that three observational learning models (live, verbal, and symbolic) significantly influence perceptions of the police. In particular, the symbolic model is significant regardless of students’ direct and indirect contact experiences with the police.Originality/value– This study is the first to examine the modeling effects on attitudes toward the police applying the classic social learning theory developed by Albert Bandura. The results highlight the importance of indirect police contact experiences in shaping young citizens’ perceptions of the police.