Weighted ensemble model for image classification

Iqball, Talib; Wani, M. Arif

doi:10.1007/s41870-022-01149-8

Cited by 17 publications

(2 citation statements)

References 29 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…While unweighted averaging is a reasonable approach when the performance of each DNN model is roughly the same, it is not optimal when the base learners consist of heterogeneous models [11]. There are several existing weighting methods for base learners including approaches based on Bayesian theory [24]- [26], the performance of each base learner [6], [7], [27], [28], and the utilization of a meta-learner [29], [30].…”

Section: ) Weighted Averagingmentioning

confidence: 99%

Toward Improving Ensemble-Based Collaborative Inference at the Edge

Kumazawa,

Yu,

Kawamura

et al. 2024

IEEE Access

View full text Add to dashboard Cite

Ensemble-based collaborative inference systems, Edge Ensembles, are deep learning edge inference systems that enhance accuracy by aggregating predictions from models deployed on each device. They offer several advantages, including scalability based on task complexity and decentralized functionality without dependency on centralized servers. In general, ensemble methods effectively improve the accuracy of deep learning, and conventional research uses several model integration techniques for deep learning ensembles. Some of these existing integration methods are more effective than those used in previous Edge Ensembles. However, it remains uncertain whether these methods can be directly applied in the context of cooperative inference systems involving multiple edge devices. This study investigates the effectiveness of conventional model integration techniques, including cascade, weighted averaging, and testtime augmentation (TTA), when applied to Edge Ensembles to enhance their performance. Furthermore, we propose enhancements of these techniques tailored for Edge Ensembles. The cascade reduces the number of models required for inference but worsens latency by sequential inference processing. To address this latency issue, we propose m-parallel cascade, which adjusts the number of models processed simultaneously to m. We also propose learning TTA policies and weights for weighted averaging using ensemble prediction labels instead of ground truth labels. In the experiments, we verified the effectiveness of each technique for Edge Ensembles. The proposed m-parallel cascade achieved a 2.8 times reduction in latency compared to the conventional cascade, even with a 1.06 times increase in computational costs. Additionally, the ensemble label-based learning demonstrated comparable effectiveness to the approach using ground truth labels.INDEX TERMS Ensemble, edge computing, collaborative inference, neural networks, cascade, test time augmentation.

show abstract

Section: ) Weighted Averagingmentioning

confidence: 99%

Toward Improving Ensemble-Based Collaborative Inference at the Edge

Kumazawa,

Yu,

Kawamura

et al. 2024

IEEE Access

View full text Add to dashboard Cite

show abstract

“…The data-driven approach for fault detection and isolation proposed in this paper was two-fold. The first part was an ensemble of classification algorithms [26] for detecting faulty condition scenarios. A range of classification models, including logistic regression [27], decision tree [28], random forest [29], Gaussian naive Bayes [30], K-nearest neighbours [31], support vector machine [32], gradient boosting [33], and AdaBoost [34], were instantiated.…”

Section: Data-driven Approachmentioning

confidence: 99%

Data-Quality Assessment for Digital Twins Targeting Multi-Component Degradation in Industrial Internet of Things (IIoT)-Enabled Smart Infrastructure Systems

Barimah,

Niculita,

McGlinchey

et al. 2023

Applied Sciences

View full text Add to dashboard Cite

In the development of analytics for PHM applications, a lot of emphasis has been placed on data transformation for optimal model development without enough consideration for the repeatability of the measurement systems producing the data. This paper explores the relationship between data quality, defined as the measurement system analysis (MSA) process, and the performance of fault detection and isolation (FDI) algorithms within smart infrastructure systems. This research employs a comprehensive methodology, starting with an MSA process for data-quality evaluation and leading to the development and evaluation of fault detection and isolation (FDI) algorithms. During the MSA phase, the repeatability of a water distribution system’s measurement system is examined to characterise variations within the system. A data-quality process is defined to gauge data quality. Synthetic data are introduced with varying data-quality levels to investigate their impact on FDI algorithm development. Key findings reveal the complex relationship between data quality and FDI algorithm performance. Synthetic data, even with lower quality, can improve the performance of statistical process control (SPC) models, whereas data-driven approaches benefit from high-quality datasets. The study underscores the importance of customising FDI algorithms based on data quality. A framework for instantiating the MSA process for IIoT applications is also suggested. By bridging data-quality assessment with data-driven FDI, this research contributes to the design of digital twins for IIoT-enabled smart infrastructure systems. Further research on the practical implementation of the MSA process for edge analytics for PHM applications will be considered as part of our future research.

show abstract

Artificial intelligence applications in education: Natural language processing in detecting misconceptions

Kökver,

Pektaş,

Çelik

2024

Educ Inf Technol

View full text Add to dashboard Cite

This study aims to determine the misconceptions of teacher candidates about the greenhouse effect concept by using Artificial Intelligence (AI) algorithm instead of human experts. The Knowledge Discovery from Data (KDD) process model was preferred in the study where the Analyse, Design, Develop, Implement, Evaluate (ADDIE) instructional design cycle was used. The dataset obtained from 402 teacher candidates was analysed by Natural Language Processing (NLP) methods. Data was classified using Machine Learning (ML), one of the AI tools, and supervised learning algorithms. It was concluded that 175 teacher candidates did not have sufficient knowledge about the concept of greenhouse effect. It was found that the AI algorithm with the highest accuracy rate and used to predict teacher candidates’ misconceptions was Multilayer Perceptron (MLP). Furthermore, through the Enhanced Ensemble Model Architecture developed by researchers, the combination of ML algorithms has achieved the highest accuracy rate. The kappa (κ) value was examined in determining the significant difference between the AI algorithm and the human expert evaluation, and it was found that there was a significant difference, and the strength of agreement was significant according to the research findings. The findings of the current study represent a significant alternative to the prevailing pedagogical approach, which has increasingly come to rely on information technologies in the process of improving conceptual understanding through the detection of conceptual misconceptions. In addition, recommendations were made for future studies.

show abstract

Weighted ensemble model for image classification

Cited by 17 publications

References 29 publications

Toward Improving Ensemble-Based Collaborative Inference at the Edge

Toward Improving Ensemble-Based Collaborative Inference at the Edge

Data-Quality Assessment for Digital Twins Targeting Multi-Component Degradation in Industrial Internet of Things (IIoT)-Enabled Smart Infrastructure Systems

Artificial intelligence applications in education: Natural language processing in detecting misconceptions

Contact Info

Product

Resources

About