Model-wise and point-wise random sample consensus for robust regression and outlier detection.

Area of Science:

• Machine Learning
• Computer Vision
• Data Science

Background:

• Traditional regression methods struggle with data outliers.
• Existing robust M-estimators offer limited protection against outliers.
• Random Sample Consensus (RANSAC) is a robust framework but has limitations.

Purpose of the Study:

• To develop advanced robust regression and outlier detection techniques.
• To introduce a novel point-wise RANSAC strategy.
• To enhance the robustness and efficiency of regression algorithms in the presence of outliers.

Main Methods:

• Development of new estimation algorithms based on model-wise and point-wise RANSAC.
• Theoretical analysis of algorithm robustness and breakdown points in a probabilistic setting.
• Application to multilayered feed-forward neural networks for regression tasks.
• Evaluation on synthetic and real-world data with high outlier contamination.
• Development and assessment of parallel implementations for improved performance.

Main Results:

• Proposed RANSAC algorithms demonstrate superior robustness against outliers compared to existing methods.
• The point-wise RANSAC strategy offers significant advantages in outlier detection.
• Algorithms show strong performance on both synthetic and real-world datasets.
• Parallel implementations effectively reduce computation time.

Conclusions:

• The novel point-wise RANSAC framework provides a powerful approach for robust regression and outlier detection.
• The developed algorithms are effective, particularly for neural network-based regression.
• The research offers practical solutions for handling noisy datasets in machine learning.