Learning Speed Research Articles

Reward estimation with scheduled knowledge distillation for dialogue policy learning

Formulating dialogue policy as a reinforcement learning (RL) task enables a dialogue system to act optimally by interacting with humans. However, typical RL-based methods normally suffer from challenges such as sparse and delayed reward problems. Besides, with user goal unavailable in real scenarios, the reward estimator is unable to generate reward reflecting action validity and task completion. Those issues may slow down and degrade the policy learning significantly. In this paper, we present a novel scheduled knowledge distillation framework for dialogue policy learning, which trains a compact student reward estimator by distilling the prior knowledge of user goals from a large teacher model. To further improve the stability of dialogue policy learning, we propose to leverage self-paced learning to arrange meaningful training order for the student reward estimator. Comprehensive experiments on Microsoft Dialogue Challenge and MultiWOZ datasets indicate that our approach significantly accelerates the learning speed, and the task-completion success rate can be improved from 0.47%∼9.01% compared with several strong baselines.

Factors Affecting Children’s Language Development: A Systematic Review

Language development is a crucial part of preschoolers’ entire growth trajectory. Meanwhile, with the development in economics and the advance in science and technology, the issues of education for preschool children have received unprecedented attention. The review focuses on the factors that may affect children’s language development, like learning speed differences, vocabulary variations, and grammatical accuracy. Therefore, this review takes a lot of reading on literature with the aim of collecting opinions and data from different perspectives, also analyzing them. Ultimately summarizing the possible factors as comprehensively as possible. The research points out that there are various factors related to children’s language development, which can be divided into internal and external factors. Internal factors include gender difference and language aptitude. The external actors, for instance, parent-infant interactions, bilingual family effects, and speech heard at school. What’s more, whether have a preference for using symbols to express needs and wants in infancy has also influenced children’s language development to a certain extent. The possibility of these results occurring is discussed in light of the separating situations. Some paradoxes of the research are also mentioned in the end.

An Enhanced Expectation Maximization Text Document Clustering Algorithm for E-Content Analysis

Nowadays, there are many types of digital materials that can be used in the classroom. Students and scholars are migrating from textbooks to digital study materials because textbooks are too large and expensive. Teachers and college students can use and modify the materials that are freely available or with some constraints for their learning and teaching. E-content can be designed, evolved, utilized, re-used, and distributed electronically from anywhere at anytime. Because of the flexibility of time, place, and speed of learning, e-content is becoming extremely popular. It can be readily and instantly shared and communicated with an infinite number of clients all across the globe. Document clustering is most commonly used to group documents that are related to a specific topic. Text document clustering can be used to group a collection of documents regarding the information they include and to deliver search results when a user searches the internet. In this paper mainly focuses on text document clustering to cope with massive collection of E-Content documents. Enhanced Expectation Maximization Text Document Clustering (EEMTDC) clustering algorithm was proposed and compared with Expectation Maximization (EM) clustering, K-Means clustering, and Hierarchical clustering (HC) algorithms. The experiment shows that the performance of proposed EEMTDC algorithm produces greater clustering accuracy than existing clustering algorithms.

A DRL-Based Intelligent Jamming Approach for Joint Channel and Power Optimization

The traditional jamming methods mainly focus on the optimization of a single domain, such as frequency domain. However, in the practical wireless communication, the jamming methods in single dimension domain are limited and difficult to cope with multidomain scenarios. To overcome these issues, this paper investigates the problem of joint decision-making for jamming channel and power in a dynamic spectrum environment. Firstly, the Markov decision process (MDP) is used to formulate the jamming channel and power joint selection problem. Then, a deep reinforcement learning- (DRL-) based jamming algorithm is proposed with the function of parallel learning and joint decision-making. Specially, to accelerate the learning speed of the algorithm, the prioritized experience reply (PER) technology is introduced. Finally, a practical jamming testbed is built to evaluate the proposed algorithm. The simulation results and the testbed results demonstrate the effectiveness of the proposed algorithm, which can guarantee the jamming effect and maximize the utilization of jamming resources.

Using binary-stiffness beams within mechanical neural-network metamaterials to learn

This work introduces the concept of applying binary-stiffness beams within a lattice to achieve a mechanical neural-network (MNN) metamaterial that learns its behaviors and properties with prolonged exposure to unanticipated ambient loading scenarios. Applying such beams to MNN metamaterials greatly increases their learning speed and simplifies the actuation demands, control circuitry, and optimization algorithms required by previously proposed concepts. A binary-stiffness beam design is proposed that uses principles of constraint manipulation and stiffness cancelation to achieve two switchable and discrete states of stiffness (i.e. binary stiffness) along its axis. The beam achieves a near-zero low-stiffness state and a large difference in stiffness between its high and low-stiffness states, which are both shown to be desirable attributes for learning mechanical behaviors. Simulations are conducted to characterize the effect of lattice size, the difference in stiffness between the constituent beam’s high and low-stiffness states, the magnitude of its low-stiffness state, and the number of simultaneously learned behaviors on MNN learning using binary-stiffness beams. Thus, this work provides a necessary step toward enabling practical artificial intelligent metamaterials.

Ion Intercalation Enabled Tunable Frequency Response in Lithium Niobite Memristors

Memristors have emerged as a viable component for developing neuromorphic hardware platforms, which can compete with biological systems in density, accuracy, and energy efficiency. Among contemporary memristive systems, intercalation-driven lithium niobite (LiNbO2) memristors have the advantage of inherent flux-driven large analog conductance tunability ( <inline-formula xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink"> <tex-math notation="LaTeX">$\Delta {R}/{R} &gt;89$ </tex-math></inline-formula> ), a wide resistance range (~10–1E <inline-formula xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink"> <tex-math notation="LaTeX">$7 \Omega $ </tex-math></inline-formula> ) selected via geometry selection, and low-power (~100–150 mV) neuromorphic functionality resembling synaptic weight updates in the synaptic membrane. Emerging neural systems require nonstatic temporal responses to implement temporally diverse architectures, such as recurrent neural networks (RNNs), yet these temporally diverse memristors are rare. There is a gap in understanding of the frequency response of nonvolatile memristors, which is a fundamental characteristic in memristive systems and ultimately dictates the speed of adaptive learning in deployed neural networks and the memory windows available for designers in RNNs. Using both large and small signal characterization methods, these memristors demonstrate up to a decade of span in tunable frequency-dependent response, statically controlled via device geometry and scaling design rules, and dynamically tuned via channel lithium concentration. Thus, the tunable frequency dependence of resistance modulation in LiNbO2 memristors is evaluated, enabling future neural network design rules to be identified for scalable multifunctional memristive systems for neuromorphic computing applications.

Theta oscillations in anterior cingulate cortex and orbitofrontal cortex differentially modulate accuracy and speed in flexible reward learning.

Flexible reward learning relies on frontal cortex, with substantial evidence indicating that anterior cingulate cortex (ACC) and orbitofrontal cortex (OFC) subregions play important roles. Recent studies in both rat and macaque suggest theta oscillations (5-10 Hz) may be a spectral signature that coordinates this learning. However, network-level interactions between ACC and OFC in flexible learning remain unclear. We investigated the learning of stimulus-reward associations using a combination of simultaneous in vivo electrophysiology in dorsal ACC and ventral OFC, partnered with bilateral inhibitory DREADDs in ACC. In freely behaving male and female rats and using a within-subject design, we examined accuracy and speed of response across distinct and precisely defined trial epochs during initial visual discrimination learning and subsequent reversal of stimulus-reward contingencies. Following ACC inhibition, there was a propensity for random responding in early reversal learning, with correct vs. incorrect trials distinguished only from OFC, not ACC, theta power differences in the reversal phase. ACC inhibition also hastened incorrect choices during reversal. This same pattern of change in accuracy and speed was not observed in viral control animals. Thus, characteristics of impaired reversal learning following ACC inhibition are poor deliberation and weak theta signaling of accuracy in this region. The present results also point to OFC theta oscillations as a prominent feature of reversal learning, unperturbed by ACC inhibition.

An enhancement of transformer-based architecture with randomized regularization for wind speed prediction

The wind power is considered as a potential renewable energy resource which requires less management cost and effort than the others like as tidal, geothermal, etc. However, the natural randomization and volatility aspects of wind in different regions have brought several challenges for efficiently as well as reliably operating the wind-based power supply grid. Thus, it is necessary to have centralized monitoring centers for managing as well as optimizing the performance of wind power farms. Among different management task, wind speed prediction is considered as an important task which directly support for further wind-based power supply resource planning/optimization, hence towards power shortage risk and operating cost reductions. Normally, considering as traditional time-series based prediction problem, most of previous deep learning-based models have demonstrated significant improvement in accuracy performance of wind speed prediction problem. However, most of recurrent neural network (RNN) as well as sequential auto-encoding (AE) based architectures still suffered several limitations related to the capability of sufficient preserving the spatiotemporal and long-range time dependent information of complex time-series based wind datasets. Moreover, previous RNN-based wind speed predictive models also perform poor prediction results within high-complex/noised time-series based wind speed datasets. Thus, in order to overcome these limitations, in this paper we proposed a novel integrated convolutional neural network (CNN)-based spatiotemporal randomization mechanism with transformer-based architecture for wind speed prediction problem, called as: RTrans-WP. Within our RTrans-WP model, we integrated the deep neural encoding component with a randomized CNN learning mechanism to softy align temporal feature within the long-range time-dependent learning context. The utilization of randomized CNN component at the data encoding part also enables to reduce noises and time-series based observation uncertainties which are occurred during the data representation learning and wind speed prediction-driven fine-tuning processes.

Machine-Learning-Based Carbon Dioxide Concentration Prediction for Hybrid Vehicles.

The current understanding of CO2 emission concentrations in hybrid vehicles (HVs) is limited, due to the complexity of the constant changes in their power-train sources. This study aims to address this problem by examining the accuracy, speed and size of traditional and advanced machine learning (ML) models for predicting CO2 emissions in HVs. A new long short-term memory (LSTM)-based model called UWS-LSTM has been developed to overcome the deficiencies of existing models. The dataset collected includes more than 20 parameters, and an extensive input feature optimization has been conducted to determine the most effective parameters. The results indicate that the UWS-LSTM model outperforms traditional ML and artificial neural network (ANN)-based models by achieving 97.5% accuracy. Furthermore, to demonstrate the efficiency of the proposed model, the CO2-concentration predictor has been implemented in a low-powered IoT device embedded in a commercial HV, resulting in rapid predictions with an average latency of 21.64 ms per prediction. The proposed algorithm is fast, accurate and computationally efficient, and it is anticipated that it will make a significant contribution to the field of smart vehicle applications.

Early childhood learning analytics: A case study of Learning Jungle

The benefits of participating in high-quality Early Childhood Education (ECE) have been recognized by people for many years; and the need for high-quality ECE has never been greater. In this case study, we focus on whether ECE can improve learning speed in five domains: social, emotional, communication, cognition, and physical development. The initial ages for each of these five domains, in months since birth, are collected and compared with that of common children as described in Nipissing District Developmental Screen (NDDS). We find that children in the ECE program learned faster with a p-value no &amp;gt;0.0078. In addition, students in an ECE program are labeled by their ages at enrollment as Cohort 1 (infant) and Cohort 2 (toddler), and we conduct the following statistical tests on their difference: Welch's t-test, Hoteling's T2-test, and survival analysis. We find that the average initial observation age of Cohort 1 is 4.82 months earlier than that of Cohort 2 with a p-value no &amp;gt;0.009. We are convinced that ECE programs could advance students' learning in all five domains.

Classroom as Community

The class is a room that is included in the school building, which functions as a place for face-to-face activities in the process of teaching and learning activities. In a class there are students who differ in terms of knowledge, learning speed, and learning styles. They differ in ethnicity, culture, social class and gender. These differences can have important implications for school teaching, curriculum, policies and practices. This is what makes the teacher able to accept, accommodate, and appreciate the diversity of students in a community in the classroom to achieve the same goal. An educator must create a positive classroom environment where all students feel welcome and are involved in classroom learning. Community is one important aspect that must exist in every class. Community can be interpreted as a group consisting of teachers and students in the classroom and interact in teaching and learning activities. An effective teacher will strive so that in the learning he carries out an effective community is formed as well. The community that exists in a class in a learning activity will greatly influence student involvement in the learning process, and ultimately the achievement of learning objectives. The concept of this community is a learning result obtained from sharing between students and teachers, students and students about a material.

Neuroadaptive global sliding mode control for nonlinear systems and its application

AbstractOne type of adaptive controller on the basis of an original dual‐loop recursive fuzzy neural network (DRFNN) with two closed loop structures for a kind of nonlinear dynamic systems is designed in this paper. Though the conventional FNN could approximate arbitrary smooth functions, the inner network parameters need to be set up in advance. Usually, the setting of the base width and central vector are lack of theoretical guidance to some degree and need to be debugged many times. However, the initial values of the center vector and the base width can be arbitrarily set and stable to optimal ones in the light of adaptive mechanism of the proposed dual recursive FNN. Besides, the dynamic dual recursive FNN possesses the ability of storing more beneficial information through constructing signal back loops, meanwhile accuracy of the approximation is higher than the traditional FNN. Furthermore, the learning speed and detection accuracy can be improved due to the fact that expert knowledge is a priori knowledge in the network structure by including if‐then rules of this FNN. To validate effectiveness of the developed scheme, simulations and three suits of experiments are carried out on a set of active power filter to demonstrate that this presented global sliding mode control using DRFNN can achieve expected performance. Eventually, some comparisons among FNN and the proposed DRFNN are implemented to suggest the dual recursive FNN can obtain more superior properties.

Extreme learning machine–radial basis function neural network–based state-of-charge estimation of lithium-ion batteries assisted with fiber Bragg grating sensor measurements

Lithium-ion batteries have been widely used in power systems for accepting large-scale renewable energy generations and in the transport sector to facilitate vehicle decarbonization, the accuracy of state-of-charge (SOC) estimation is extremely crucial for the battery systems to operate safely and reliably. Many approaches have been developed for battery SOC estimation based on the measurable electrical signals. However, the electrical measurements are usually corrupted by sensor noise and are also limited by insulation, corrosion, and electromagnetic compatibility. To address these problems, fiber Bragg grating (FBG) sensors are used to replace thermocouple sensors, the battery SOC therefore can be estimated based on the measurement of the strain information. In order to train the SOC estimation model for lithium-ion batteries, an FBG sensor–based battery SOC estimator, namely, ELM-RBFNN, is developed using the radial basis function neural network (RBFNN) identified by extreme learning machine (ELM). According to the ELM principle, the ELM-RBFNN model randomly selects the radial basis function (RBF) centers and determines the parameters of the hidden-layer nodes without any manual adjustment. The ELM-RBFNN estimation model has the advantages of fast learning speed, global optimization, and excellent generalization. Experimental results confirm that compared to the linear estimator and the polynomial estimator, the developed SOC estimator is able to produce more accurate results.

Deterministic learning-based neural network control with adaptive phase compensation

Under the persistent excitation (PE) condition, the real dynamics of the nonlinear system can be obtained through the deterministic learning-based radial basis function neural network (RBFNN) control. However, in this scheme, the learning speed and accuracy are limited by the tradeoff between the PE levels and the approximation capabilities of the neural network (NN). Inspired by the frequency domain phase compensation of linear time-invariant (LTI) systems, this paper presents an adaptive phase compensator employing the pure time delay to improve the performance of the deterministic learning-based adaptive feedforward control with the reference input known a priori. When the adaptive phase compensation is applied to the hidden layer of the RBFNN, the nonlinear approximation capability of the RBFNN is effectively improved such that both the learning performance (learning speed and accuracy) and the control performance of the deterministic learning-based control scheme are improved. Theoretical analysis is conducted to prove the stability of the proposed learning control scheme for a class of systems which are affine in the control. Simulation studies demonstrate the effectiveness of the proposed phase compensation method.

A deep ensemble learning method for single finger-vein identification.

Finger-vein biometrics has been extensively investigated for personal verification. Single sample per person (SSPP) finger-vein recognition is one of the open issues in finger-vein recognition. Despite recent advances in deep neural networks for finger-vein recognition, current approaches depend on a large number of training data. However, they lack the robustness of extracting robust and discriminative finger-vein features from a single training image sample. A deep ensemble learning method is proposed to solve the SSPP finger-vein recognition in this article. In the proposed method, multiple feature maps were generated from an input finger-vein image, based on various independent deep learning-based classifiers. A shared learning scheme is investigated among classifiers to improve their feature representation captivity. The learning speed of weak classifiers is also adjusted to achieve the simultaneously best performance. A deep learning model is proposed by an ensemble of all these adjusted classifiers. The proposed method is tested with two public finger vein databases. The result shows that the proposed approach has a distinct advantage over all the other tested popular solutions for the SSPP problem.

Feasibility of serial neurocognitive assessment using Cogstate during and after therapy for childhood leukemia.

Neurocognitive impairment is frequently observed among survivors of childhood acute lymphoblastic leukemia (ALL) within the domains of attention, working memory, processing speed, executive functioning, and learning and memory. However, few studies have characterized the trajectory of treatment-induced changes in neurocognitive function beginning in the first months of treatment, to test whether early changes predict impairment among survivors. If correct, we hypothesize that those children who are most susceptible to early impairment would be ideal subjects for clinical trials testing interventions designed to protect against treatment-related neurocognitive decline. In this pilot study, we prospectively assessed neurocognitive functioning (attention, working memory, executive function, visual learning, and processing speed), using the Cogstate computerized battery at six time points during the 2years of chemotherapy treatment and 1-year post-treatment (Dana-Farber Cancer Institute ALL Consortium protocol 11-001; NCT01574274). Forty-three patients with ALL consented to serial neurocognitive testing. Of the 31 participants who remained on study through the final time point, 1year after completion of chemotherapy, 28 (90%) completed at least five of six planned Cogstate testing time points. Performance and completion checks indicated a high tolerability (≥ 88%) for all subtests. One year after completion of treatment, 10 of 29 patients (34%) exhibited neurocognitive function more than 2 standard deviations below age-matched norms on one or more Cogstate subtests. Serial collection of neurocognitive data (within a month of diagnosis with ALL, during therapy, and 1-year post-treatment) is feasible and can be informative for evaluating treatment-related neurocognitive impairment.

Exploring the Benefits and Challenges of AI-Language Learning Tools

AI language learning tools are computer programs or software applications that use artificial intelligence (AI) algorithms to help users learn and improve their skills in a foreign language. These technologies provide a variety of advantages, including the capacity to reduce time and increase learning speed, to give learners personalized learning experiences, and to introduce them to other cultures. In this study, scientific literature has been reviewed in order to analyze opportunities, challenges and limitations to using AI language learning tools, including the need for more human interaction, and contextual nuances of language, and dependence on large amounts of data for training. &#x0D; These AI tools have several potential future advances, including integrating VR and AR technologies, improving natural language processing algorithms, and developing more advanced adaptive learning algorithms. By addressing these issues and constraints, AI learning systems have the potential to become even more powerful and impactful tools for language learning, and their integration plays a crucial role in designing more effective and efficient learning user experience solutions.&#x0D;

Multi-agent deep reinforcement learning based distributed control architecture for interconnected multi-energy microgrid energy management and optimization

Environmental and climate change concerns are pushing the rapid development of new energy resources (DERs). The Energy Internet (EI), with the power-sharing functionality introduced by energy routers (ERs), offers an appealing alternative for DER systems. However, previous centralized control schemes for EI systems that follow a top-down architecture are unreliable for future power systems. This study proposes a distributed control scheme for bottom-up EI architecture. Second, model-based distributed control methods are not sufficiently flexible to deal with the complex uncertainties associated with multi-energy demands and DERs. A novel model-free/data-driven multiagent deep reinforcement learning (MADRL) method is proposed to learn the optimal operation strategy for the bottom-layer microgrid (MG) cluster. Unlike existing single-agent deep reinforcement learning methods that rely on homogeneous MG settings, the proposed MADRL adopts a form of decentralized execution, in which agents operate independently to meet local customized energy demands while preserving privacy. Third, an attention mechanism is added to the centralized critic, which can effectively accelerate the learning speed. Considering the bottom-layer power exchange request and the predicted electricity price, model predictive control of the upper layer determines the optimal power dispatching between the ERs and main grid. Simulations with other alternatives demonstrate the effectiveness of the proposed control scheme.

DETECTION OF NETWORK INTRUSIONS USING MACHINE LEARNING ALGORITHMS AND FUZZY LOGIC

The study proposed a new method of constructing a set of important features for solving classification problems. This method is based on the idea of using an ensemble of estimators of the importance of features with summarization and the final result of the ensemble with the help of fuzzy logic algorithms. Statistical criteria (chi2, f_classif, correlation coefficient), mean decrease in impurity (MDI), mutual information criterion (mutual_info_classif) were used as estimators of the importance of features. Reducing the number of features on all data sets affects the accuracy of the assessment according to the criterion of the average reduction of classification errors. As long as the group of features in the data set for training contains the first features with the greatest influence, the accuracy of the model is at the initial level, but when at least one of the features with a large impact is excluded from the model, the accuracy of the model is noticeably reduced. The best classification results for all studied data sets were provided by classifiers based on trees or nearest neighbors: DesignTreeClassifier, ExtraTreeClassifier, KNeighborsClassifier. Due to the exclusion of non-essential features from the model, a noticeable increase in the speed of learning is achieved (up to 60-70%). Ensemble learning was used to increase the accuracy of the assessment. The VotingClassifier classifier, built on the basis of algorithms with the maximum learning speed, provided the best learning speed indicators. For future work, the goal is to further improve the proposed IDS model in the direction of improving the selection of classifiers to obtain optimal results, and setting the parameters of the selected classifiers, improving the strategy of generalizing the results of individual classifiers. For the proposed model, the ability to detect individual types of attacks with multi-class prediction is of significant interest.

A Novel Model-Assisted Decentralized Multi-Agent Reinforcement Learning for Joint Optimization of Hybrid Beamforming in Massive MIMO mmWave Systems

To meet the explosive increasing demands for high data rates, hybrid beamforming has become a promising paradigm for massive MIMO millimeter wave (mmWave) systems. In this paper, we study the joint optimization of hybrid beamforming for maximizing the achievable sum rate in networked dynamic massive MIMO mmWave systems. To address this problem, we propose a novel Model-Assisted Decentralized Multi-Agent Reinforcement Learning (MAD-MARL) algorithm, which incorporates the following three designs: 1) a model-based prediction approach, which can improve the policy learning speed greatly and also predict each agent's potential future so as to alleviate the impact of delayed information sharing in a networked system, 2) an attentional prediction sharing approach, which is to effectively reduce the non-stationarity of the environment by combining the proposed model-based prediction and self-attention strategy for reaching global consensus and cooperative inter-agent coordination in a distributed manner, and 3) a decentralized model-free learning approach, which trains the agents with the assistance of model-based predictions. Extensive simulations are conducted and the numerical results demonstrate that MAD-MARL can significantly increase the learning speed and also greatly improve the overall performance compared with existing work.