Learning System Research Articles

DREAM: Debugging and Repairing AutoML Pipelines

Deep Learning models have become an integrated component of modern software systems. In response to the challenge of model design, researchers proposed Automated Machine Learning (AutoML) systems, which automatically search for model architecture and hyperparameters for a given task. Like other software systems, existing AutoML systems have shortcomings in their design. We identify two common and severe shortcomings in AutoML, performance issue (i.e., searching for the desired model takes an unreasonably long time) and ineffective search issue (i.e., AutoML systems are not able to find an accurate enough model). After analyzing the workflow of AutoML, we observe that existing AutoML systems overlook potential opportunities in search space, search method, and search feedback, which results in performance and ineffective search issues. Based on our analysis, we design and implement DREAM, an automatic and general-purpose tool to alleviate and repair the shortcomings of AutoML pipelines and conduct effective model searches for diverse tasks. It monitors the process of AutoML to collect detailed feedback and automatically repairs shortcomings by expanding search space and leveraging a feedback-driven search strategy. Our evaluation results show that DREAM can be applied on two state-of-the-art AutoML pipelines and effectively and efficiently repair their shortcomings.

Automated acute skin toxicity scoring in a mouse model through deep learning.

This study presents a novel approach to skin toxicity assessment in preclinical radiotherapy trials through an advanced imaging setup and deep learning. Skin reactions, commonly associated with undesirable side effects in radiotherapy, were meticulously evaluated in 160 mice across four studies. A comprehensive dataset containing 7542 images was derived from proton/electron trials with matched manual scoring of the acute toxicity on the right hind leg, which was the target area irradiated in the trials. This dataset was the foundation for the subsequent model training. The two-step deep learning framework incorporated an object detection model for hind leg detection and a classification model for toxicity classification. An observer study involving five experts and the deep learning model, was conducted to analyze the retrospective capabilities and inter-observer variations. The results revealed that the hind leg object detection model exhibited a robust performance, achieving an accuracy of almost 99%. Subsequently, the classification model demonstrated an overall accuracy of about 85%, revealing nuanced challenges in specific toxicity grades. The observer study highlighted high inter-observer agreement and showcased the model's superiority in accuracy and misclassification distance. In conclusion, this study signifies an advancement in objective and reproducible skin toxicity assessment. The imaging and deep learning system not only allows for retrospective toxicity scoring, but also presents a potential for minimizing inter-observer variation and evaluation times, addressing critical gaps in manual scoring methodologies. Future recommendations include refining the system through an expanded training dataset, paving the way for its deployment in preclinical research and radiotherapy trials.

Enhanced Magnetic Resonance Imaging-Based Brain Tumor Classification with a Hybrid Swin Transformer and ResNet50V2 Model

Brain tumors can be serious; consequently, rapid and accurate detection is crucial. Nevertheless, a variety of obstacles, such as poor imaging resolution, doubts over the accuracy of data, a lack of diverse tumor classes and stages, and the possibility of misunderstanding, present challenges to achieve an accurate and final diagnosis. Effective brain cancer detection is crucial for patients’ safety and health. Deep learning systems provide the capability to assist radiologists in quickly and accurately detecting diagnoses. This study presents an innovative deep learning approach that utilizes the Swin Transformer. The suggested method entails integrating the Swin Transformer with the pretrained deep learning model Resnet50V2, called (SwT+Resnet50V2). The objective of this modification is to decrease memory utilization, enhance classification accuracy, and reduce training complexity. The self-attention mechanism of the Swin Transformer identifies distant relationships and captures the overall context. Resnet 50V2 improves both accuracy and training speed by extracting adaptive features from the Swin Transformer’s dependencies. We evaluate the proposed framework using two publicly accessible brain magnetic resonance imaging (MRI) datasets, each including two and four distinct classes, respectively. Employing data augmentation and transfer learning techniques enhances model performance, leading to more dependable and cost-effective training. The suggested model achieves an impressive accuracy of 99.9% on the binary-labeled dataset and 96.8% on the four-labeled dataset, outperforming the VGG16, MobileNetV2, Resnet50V2, EfficientNetV2B3, ConvNeXtTiny, and convolutional neural network (CNN) algorithms used for comparison. This demonstrates that the Swin transducer, when combined with Resnet50V2, is capable of accurately diagnosing brain tumors. This method leverages the combination of SwT+Resnet50V2 to create an innovative diagnostic tool. Radiologists have the potential to accelerate and improve the detection of brain tumors, leading to improved patient outcomes and reduced risks.

Comparison and AI-based prediction of graph comprehension skills based on the visual strategies of first-year physics and medicine students

Graphical representations of data are common in many disciplines. Previous research has found that physics students appear to have better graph comprehension skills than students from social science disciplines, regardless of the task context. However, the graph comprehension skills of physics students have not yet been compared with (veterinary) medicine students, both of which are disciplines that require multiple science, technology, engineering, and mathematics (STEM) courses. This study extends previous research on this subject by exploring whether physics majors possess superior graph comprehension skills due to their study discipline. Here, participants solved 24 graph comprehension tasks across various subjects, including mathematics, physics, and medicine; these tests were conducted at the beginning and end of their first semester. Graph comprehension gain was calculated based on the percentage of correct and incorrect answers in the pretest and the post-test. In addition to these comparisons, we replicated previous research that successfully distinguished correct and incorrect solvers based on their visual behavior by using a novel machine-learning method tailored to small datasets. Through this replication of statistical analyses, we aim to ensure the reliability of adaptive learning systems in the future, regardless of data size, using the same machine-learning method. Physics and medical students were found to exhibit relatively similar graph comprehension gain; this is in contrast to previous research comparing physics and non-STEM students. Our results also revealed that both physics and medical students use similar visual strategies to solve these tasks. However, correct and incorrect solvers could be distinguished via machine-learning methods regardless of their discipline. Our research suggests that visual behavior is a good predictor of graph comprehension skills. Published by the American Physical Society 2024

An insight of undergraduate students’ attitude and satisfaction with online home-based learning system

The purpose of this study is to investigate different factors associated with remote online home-based learning (thereafter named OHL), including technical system quality, perceived quality of contents, perceived ease of use, and perceived usefulness in relation to the satisfaction of undergraduate students following the post-COVID-19 pandemic in Malaysia. Additionally, the mediating roles of attitude are also investigated. Two hundred questionnaires were distributed using judgmental sampling method and 156 completed responses were collected. The data were subsequently analyzed using PLS-SEM. The findings imply that the OHL system is an effective method although it is challenging to operate. In terms of perceived technical system quality, OHL is currently more gratifying for students; however, some have reported that the quality of the content delivered via the remote system is still unsatisfactory. Moreover, the study found that attitude is a significant determinant of undergraduates’ satisfaction with OHL. This study contributes to the advancement of current knowledge by inspecting the factors of the Undergraduate Level OHL System using the mediating roles of attitude. In terms of underpinning theories, Technology Acceptance Model and Information System Model were employed as the guiding principles of the current study.

Peningkatan Akses dan Peluang Kewirausahaan Mahasiswa melalui Pembelajaran Fotografi Berbasis Proyek Industri di Sistem Pembelajaran Jaringan

The program to enhance student access and entrepreneurial opportunities through industry-based photography project learning in an online learning system aims to develop students' technical and managerial skills. This program combines theory and practice through real industry-relevant projects, providing a comprehensive and interactive learning experience. Results show significant improvements in students' skills and motivation. With continued support from various parties, this program is expected to sustain improvements in education quality and entrepreneurial opportunities.

High-risk medication errors: Insight from the UK National Reporting and learning system

BackgroundEnsuring patient safety is of paramount importance in healthcare systems. Rising concerns about medical errors in the UK have necessitated a greater focus on studying the nature of such errors, particularly those involving high-risk medications. ObjectivesTo conduct a retrospective analysis of incidents related to patient safety in the UK based on data from the National Rporting and Learning System (NRLS). MethodsThis study was conducted based on a review of the National Reporting and Learning System (NRLS) patient safety reports published between January 1, 2015, and December 31, 2015. NHS Improvement provides details regarding incidents following approval using a data-sharing agreement. In total, 1500 incidents were analszed and equally divided among the three categories of high-risk drugs: opioids, insulin, and anticoagulants. Excel® features and deductive reasoning (thematic analysis) were used for data analysis. ResultsThe results showed that the insulin category had both the highest risk and most errors compared with anticoagulants and opioids. These errors primarily result from issues related to administering, prescribing, and dispensing the drugs. Inadequate drug checks, communication difficulties among staff and patients, and high staff workloads are often linked to these errors. ConclusionThis study confirms that the NRLS database is a valuable source of data, and the suggestions put forth, based on these results, could contribute to the formulation of measures that diminish the occurrence of errors related to high-risk drugs in healthcare settings. Information technology should enhance medication safety by tracking the process of medication use.

Boosting Adaptive Weighted Broad Learning System for Multi-Label Learning

Boosting Adaptive Weighted Broad Learning System for Multi-Label Learning

Enhancing Learning Management Systems (LMS) Through Kano Analysis

Abstract This article explores the critical role of information and communication technology (ICT) in modern educational settings and emphasizes the importance of understanding stakeholders’ needs in the development of digital learning tools, particularly Learning Management Systems (LMS). Drawing from the Kano Model of Satisfaction, the study examines how stakeholders evaluate digital learning products and proposes a new LMS design tailored to their preferences. The paper provides insights into the acceptance and utilization of digital learning tools by stakeholders in Higher Education Institutions (HEIs), highlighting the challenges and opportunities in incorporating LMS into educational organizations. Additionally, it presents new functional factors for newly designed LMS and discusses the implications of digital learning systems in the context of modern education. Finally, the article addresses the limitations of current LMSs and underscores the importance of stakeholder acceptance for successful implementation in educational institutions.

A Novel Four-Modal Nano-Sensor Based on Two-Dimensional Mxenes and Fully Connected Artificial Neural Networks for the Highly Sensitive and Rapid Detection of Ochratoxin A

Timely and accurate on-site detection of ochratoxin A (OTA) is extremely important for global public health. In this study, a fluorescence/colorimetric biosensor based on Ti3C2 nano-materials (Ti3C2-NMS) and a machine-learning (ML) based fluorescence/colorimetric intelligent learning system for detection of OTA concentration (COTA) were developed. The sensor was fabricated by functionalizing Ti3C2-NMS prepared by physical-exfoliation assisted metal-ion-induction using ssDNA. The Ti3C2-NMS exhibited good fluorescence quenching characteristics (FQC) and peroxidase-like activity (PLA). More surprisingly, the functionalization of Ti3C2-NMS by ssDNA further enhanced the FQC and PLA of the material, which could be used for dual-mode detection of OTA. When different COTA existed, ssDNA competitively bound to OTA, resulting in regular changes in fluorescence and colorimetric signals of the sensor, which realized the accurate and sensitive biosensing detection of OTA in two modalities. Based on a series of fluorescent/colorimetric RGB datasets collected by a self-developed application, a dual-channel ML model had been developed. This model can be integrated into mobile phones, clouds, and PCs to achieve intelligent sensing detection of OTA with the assistance of fully connected artificial neural networks. The method constructed had high specificity, low cost, and fast responsiveness, with a LOD as low as 1.58 pg mL-1, indicating excellent potential for application and promotion.

IEEE Transactions on Neural Networks and Learning Systems Information for Authors

IEEE Transactions on Neural Networks and Learning Systems Information for Authors

Motion Controller for Multi-Joint Robotic Arm with Deep Cascade Gated Bayesian Broad Learning System

Intelligent controllers based on the broad learning system can simplify the process of model parameter adjustment, finding wide applications in the motion control of multi-joint robotic arms. However, motion controllers for multi-joint robotic arms based on broad learning system exhibit insufficient precision and overlook the impact of joint motion commonalities on controller design. Therefore, this paper proposes a novel motion control strategy for a multi-joint robotic arm based on a deep cascade feature-enhancement gated Bayesian broad learning system. Firstly, the motion controller of the deep cascade feature-enhancement Bayesian broad learning system is constructed to enhance the robotic arm motion control precision. Secondly, an incremental node generation module with an attention-gated mechanism is constructed to capture the unique motion characteristics of the target joints, which is further combined with model generalization to simplify the motion control process of the multi-joint robotic arm. Finally, controller convergence is enhanced by combining it with the Lyapunov theory to constrain the learning parameters. Simulations and physical experiments are designed to verify the feasibility and superiority of the proposed motion control strategy. The results demonstrated that the strategy improved the accuracy of robotic arm motion control, with the root mean square error in position tracking reduced to 0.0019 rad. This represents a 93.39% reduction in error compared to existing techniques.

Cellular-resolution optogenetics reveals attenuation-by-suppression in visual cortical neurons

The relationship between neurons' input and spiking output is central to brain computation. Studies in vitro and in anesthetized animals suggest that nonlinearities emerge in cells' input-output (IO; activation) functions as network activity increases, yet how neurons transform inputs in vivo has been unclear. Here, we characterize cortical principal neurons' activation functions in awake mice using two-photon optogenetics. We deliver fixed inputs at the soma while neurons' activity varies with sensory stimuli. We find that responses to fixed optogenetic input are nearly unchanged as neurons are excited, reflecting a linear response regime above neurons' resting point. In contrast, responses are dramatically attenuated by suppression. This attenuation is a powerful means to filter inputs arriving to suppressed cells, privileging other inputs arriving to excited neurons. These results have two major implications. First, somatic neural activation functions in vivo accord with the activation functions used in recent machine learning systems. Second, neurons' IO functions can filter sensory inputs-not only do sensory stimuli change neurons' spiking outputs, but these changes also affect responses to input, attenuating responses to some inputs while leaving others unchanged.

SecDefender: Detecting low-quality models in multidomain federated learning systems

Federated learning (FL) is an innovative distributed learning paradigm that permits multiple parties to train models collaboratively while protecting individual privacy. However, it encounters security challenges, making it vulnerable to several adversarial attacks and leading to compromising model performance. Existing research on FL poisoning attacks and defense techniques tends to be application-specific, primarily emphasizing attack capabilities. However, it fails to consider inherent vulnerabilities in FL and the impact of attack intensity. To our knowledge, no existing work has delved into these issues within a multi-domain FL environment. This paper addresses these concerns by investigating the consequences of targeted label-flipping attacks within FL systems and comprehensively examining the effects of the attacks in single-label, double-label, and triple-label scenarios with different levels of poisoning intensities. Additionally, we investigate the influence of a temporal label-flipping attack, where we study the impact of adversaries available only for specific federated training rounds. Moreover, we propose a novel server-based defense mechanism called SecDefender to detect low-quality models in both IID and non-IID settings of multi-domain environments. Our approach is rigorously evaluated against state-of-the-art alternatives using six benchmark datasets: CIC-Darknet2020, Fashion-MNIST, FEDMNIST, GTSR, HAR, and MNIST. Extensive experiments demonstrate that our proposed SecDefender significantly enhances its performance by over 65% in terms of source class recall, maintaining a low attack success rate. Consequently, there is a 1 to 2% enhancement in global model accuracy compared to existing approaches.

IEEE Transactions on Neural Networks and Learning Systems Publication Information

IEEE Transactions on Neural Networks and Learning Systems Publication Information

Neuroplasticity Meets Artificial Intelligence: A Hippocampus-Inspired Approach to the Stability–Plasticity Dilemma

The stability–plasticity dilemma remains a critical challenge in developing artificial intelligence (AI) systems capable of continuous learning. This perspective paper presents a novel approach by drawing inspiration from the mammalian hippocampus–cortex system. We elucidate how this biological system’s ability to balance rapid learning with long-term memory retention can inspire novel AI architectures. Our analysis focuses on key mechanisms, including complementary learning systems and memory consolidation, with emphasis on recent discoveries about sharp-wave ripples and barrages of action potentials. We propose innovative AI designs incorporating dual learning rates, offline consolidation, and dynamic plasticity modulation. This interdisciplinary approach offers a framework for more adaptive AI systems while providing insights into biological learning. We present testable predictions and discuss potential implementations and implications of these biologically inspired principles. By bridging neuroscience and AI, our perspective aims to catalyze advancements in both fields, potentially revolutionizing AI capabilities while deepening our understanding of neural processes.

Developing a three-strand approach to playfulness in learning development

In this presentation, we outlined the development of a new Academic Skills Service at Oxford International Education Group, which, whilst supporting students from seven university-based international colleges, implemented a three-strand approach to playfulness in learning development. The presentation focused on three examples of playful approaches: The use of digital and traditional game-based learning within otherwise performance-driven classes. The creation of playful spaces in the form of extracurricular cross-college book and Maths movies clubs. The development of a gamified Moodle learning system through activity badging and levelling up. These initiatives reflect current research regarding the value of playful approaches in student learning and in counterbalancing a predominantly performance-driven higher education environment (Langan and Smart, 2018; Whitton, 2018; Forbes, 2021). Firstly, game-based learning can increase student enjoyment, motivation, and participation in the learning process and promote social belonging (James and Nerantzi, 2019; Forbes, 2021). Secondly, online book/Maths clubs can operate as playful spaces where student learning is supported in a positive social environment and creativity and risk-taking become part of the learning process (Whitton, 2018; James and Nerantzi, 2019; Forbes, 2021). Thirdly, although the gamified Moodle learning system relies on extrinsic rewards, which reflect the performance-led ethos of HE (Nørgård, Toft-Nielsen and Whitton, 2017), it can also increase student motivation in completing learning activities. This presentation proposed that learning development departments, whilst supporting student outcomes and the student experience, may encounter fewer performance-led constraints compared with programme-based delivery, enabling the utilisation of playful approaches that are otherwise underused in HE.

Investigation of the Process Parameters in Rotary Friction Welded Dissimilar AA7075/AA5083 Aluminum Alloy Joints on Fatigue Initiation using FEA and ANN

The rotary friction welding (RFW) method is one of the most widespread methods in the world for producing bimetallic components that require high mechanical strength. Simulations play a vital role in improving energy efficiency and reducing environmental impact, aligning with the sustainability goals of modern industry. A neural network (NN)-based incremental learning system was developed to predict crack growth and fatigue for AA5083 and AA7075 aluminum alloys. The results indicate the ability of this method to accommodate the input temperatures and the S-N curve and provide reliable predictions of expected fatigue. This method can reduce labor costs and time spent on crack propagation tests, enhancing the effectiveness of production processes and reducing process costs. This work also reveals the ability of neural . It maynetworks (NN) in monotonic function extrapolation like the S-N curve, which may pave the way for a wide variety of monotonic function-predicting problems. In future studies, a neural network (NN)-based increment learning scheme could be trained with random parts of individual S–N curves and applied to predict the rest. Additionally, the verification utilizing AISI 2205 and AISI 1020 steel has observed that neural networks may obtain S-N curve values for another metal with less than an 8% error rate. Friction pressure increases temperature, deformation, and stress in welding processes. Friction pressure 17 MPa increases temperature to 355 degrees Celsius, while Friction pressure 23 MPa increases deformation to 0.020 mm. A friction pressure of 29 MPa increases equivalent stress to 110 MPa. The indication of the S-N curve shows that increasing welding pressure increases Alternating Stress. Friction pressure also increases life, with minimum life cycles reaching 171040 cycles at 17 MPa, 195560 cycles at 23 MPa, and 283690 cycles at 29 MPa. Comparing research and simulation results, convergence is less than 8%, reducing error.

Application of Reservoir Computers for Chaos Synchronization and Cracking Chaos-Based Cryptography in Fermionic 2D Thirring and 4D Gursey Models with Spinor Fields

Reservoir computers have recently become one of the most widely exploited model-free approaches to chaos synchronization due to their fast convergence speed and simple yet effective learning rules. In this study, reservoir computing has been utilized to emulate the dynamics of chaotic Thirring and Gursey systems. In the supervised learning of the reservoir, one-step ahead states of the dynamical systems have been employed as the teaching signals. After the training phase, the reservoir was first run autonomously and then weakly driven by chaotic systems. It has been shown that the trained reservoir computers can exhibit the same characteristics as the attractors of the learned chaotic systems, which enable their use in synchronization tasks of chaotic systems with possible application in cracking chaos-based cryptography. To investigate the effect of noise on the performance of reservoir computers, noise signals with different colors and amplitudes have been included in the transmitted signals. The obtained results indicate that the proposed scheme can provide lower error metrics for both models.

Looking to see, listening to hear, learning to understand: centring Indigenous relationality to pursue a more-than-disciplinary academy

Indigenous scholars argue that the reciprocal relationality of life should be taken more seriously in scholarship responding to environmental crisis; however, much of orthodox academia analyses the environment as a separate category through the physical and social science disciplines. This article shows how human exceptionalism, human centrism, racial discrimination, and Euro-American centric privilege interweave in academic institutions and practices, often invisibly and thus insidiously, to dismiss Indigenous scholarship and lived experiences. We explore how to meaningfully address these problematics, presenting examples from our academic practices to overturn colonial and imperial privilege. We present a more-than-disciplinary approach to undiscipline nature, knowledge, and peoples. We argue that this paradigm shift is best led by Indigenous leaders in relationality, whose expert knowledge inheritance already holds nature with society. We present an Indigenous, pedological approach, to help destabilise orthodox academic disciplinary approaches and organisational structures through foregrounding Indigenous learning and knowledge systems.