Robust Extreme Learning Machine With its Application to Indoor Positioning.

Area of Science:

• Engineering
• Computer Science
• Signal Processing

Background:

• Indoor positioning systems (IPSs) are crucial for location-based services but are hindered by measurement noise.
• Existing IPS algorithms struggle to maintain performance under noisy conditions.

Purpose of the Study:

• To propose novel robust extreme learning machines (RELMs) for enhancing the performance of indoor positioning systems (IPSs).
• To address the challenge of noisy measurements in IPSs using two distinct RELM constraints: close-to-mean and small-residual.

Main Methods:

• Developed two types of RELMs: one with a close-to-mean constraint and another with a small-residual constraint.
• Formulated RELMs using second-order cone programming, offering both random-hidden-nodes and kernelized approaches.
• Investigated the computation of covariance in feature space for improved model performance.

Main Results:

• Simulations and real-world experiments demonstrated the effectiveness of the proposed RELMs.
• The RELM algorithms significantly improved accuracy and repeatability in IPS performance.
• A notable reduction in deviation and worst-case error was observed compared to baseline algorithms.

Conclusions:

• The proposed robust extreme learning machines effectively mitigate the impact of noisy measurements in indoor positioning systems.
• RELMs offer a superior alternative to existing algorithms, enhancing both accuracy and reliability for indoor localization.
• The study validates the practical applicability and performance benefits of RELMs in real-world IPS applications.