Extreme Learning Machine Applications in Classification Systems

Extreme Learning Machines (ELMs) represent a class of rapid‐training algorithms for single‐hidden‐layer feedforward neural networks that assign random weights to input and hidden layers and determine output weights via a closed‐form solution. This randomisation yields remarkable speedups over iterative backpropagation, while retaining strong generalisation performance. In classification systems, ELMs have been deployed for tasks such as handwritten digit recognition, medical image diagnosis, speech and language identification, anomaly detection and one‐class classification. Variants extend the core approach to incorporate kernel functions, deep architectures and ensemble strategies, thereby enhancing feature representation, reducing sensitivity to hyperparameters and combating overfitting. Random Vector Functional Link (RVFL) networks build on the same philosophy by introducing direct input–output connections, improving information flow and approximation power. Deep and ensemble versions of RVFL permit multilayer feature extraction without costly gradient‐based tuning and support incremental or online learning in dynamic environments. These capabilities have underpinned real‐time decision systems in industrial rolling‐force prediction, brain–computer interfaces and large‐scale image and signal processing. Emerging research emphasises integration with sparse learning, probabilistic estimation and automated hyperparameter optimisation to broaden applicability across high‐dimensional, noisy and streaming datasets.

Research from Nature Portfolio

No recent Nature Portfolio content available.

Research from all publishers

Recent work on online learning has leveraged ensemble deep Random Vector Functional Link models to address the limitations of static training. A novel architecture stacks multiple randomised layers, each producing an output layer whose aggregate forms a robust ensemble classifier. This approach enables incremental growth, closed‐form updates and enhanced stability, achieving superior classification accuracy on over 70% of benchmark datasets compared with existing randomisation‐based methods.

A comprehensive survey of randomised neural networks has charted the evolution of RVFL models, detailing shallow, deep, ensemble and ensemble‐deep variants. It highlights architectural innovations such as sparse receptive fields, direct links and skip connections inspired by residual networks, and reviews hyperparameter optimisation techniques to bolster classification generalisation. The survey identifies challenges in feature diversity maintenance and proposes future directions in adaptive randomisation and hybrid deep architectures.

An authoritative review of Extreme Learning Machines has consolidated theoretical analyses of universal approximation and generalisation bounds, while cataloguing algorithmic enhancements addressing stability, efficiency and accuracy. The review summarises diverse classification applications in medical imaging, multimedia pattern recognition and one‐class anomaly detection, and critically examines debates around the reproducibility of randomly initialised models, suggesting best practices for robust deployment.

Extreme Learning Machine Applications in Classification Systems publication trend

The graph below shows the total number of articles in extreme learning machine applications in classification systems across all publications each year (not limited to Nature Index journals).

Technical terms

Extreme Learning Machine (ELM): A learning algorithm for single-hidden-layer feedforward networks that randomises hidden node parameters and computes output weights analytically.

Single Hidden Layer Feedforward Neural Network (SLFN): A neural architecture with one layer of hidden neurons between inputs and outputs, often used for rapid approximations.

Random Vector Functional Link (RVFL): An extension of ELM that adds direct connections between input and output layers to improve information flow and universal approximation.

Ensemble Deep Random Vector Functional Link (edRVFL): A multi-layer randomised network that trains an ensemble of output layers in closed form, enhancing stability and representation without iterative backpropagation.

References

1. Online learning using deep random vector functional link network. Engineering Applications of Artificial Intelligence (2023).
2. A review on extreme learning machine. Multimedia Tools and Applications (2021).
3. Deep Extreme Learning Machine and Its Application in EEG Classification. Mathematical Problems in Engineering (2015).
4. One‐Class Classification with Extreme Learning Machine. Mathematical Problems in Engineering (2015).
5. Fast, Simple and Accurate Handwritten Digit Classification by Training Shallow Neural Network Classifiers with the ‘Extreme Learning Machine’ Algorithm. PLOS ONE (2015).
6. Spoken language identification based on the enhanced self-adjusting extreme learning machine approach. PLOS ONE (2018).
7. An Improved Kernel Based Extreme Learning Machine for Robot Execution Failures. The Scientific World JOURNAL (2014).
8. A Multiple Hidden Layers Extreme Learning Machine Method and Its Application. Mathematical Problems in Engineering (2017).
9. Improving Classification Performance through an Advanced Ensemble Based Heterogeneous Extreme Learning Machines. Computational Intelligence and Neuroscience (2017).
10. Random vector functional link network: Recent developments, applications, and future directions. Applied Soft Computing (2023).
11. Rolling Force Prediction of Hot Rolling Based on GA‐MELM. Complexity (2019).
12. Sparse Bayesian Broad Learning System for Probabilistic Estimation of Prediction. IEEE Access (2020).