Scalable Pattern Recognition In High-Dimensional Spaces Using Deep Learning Architectures

Abstract

Many problems in science and engineering can be formulated as geometric pattern recognition tasks in high-dimensional spaces. However, traditional convolution-based approaches are often unable to represent long-term dependencies and scale efficiently with increasing dimensionality. To address these drawbacks, this research proposes a scalable pattern recognition framework based on deep learning (DL) for high-dimensional data analysis. The proposed approach utilizes a High-Dimensional Dataset containing geometric pattern records extracted from image-based visual pattern data. During the preprocessing stage, noise handling is identified and handled appropriately, followed by Min-Max normalization to scale feature values and improve model convergence. Subsequently, Principal Component Analysis (PCA) is used for feature extraction and dimensionality reduction purposes. The proposed approach is first evaluated on synthetic datasets involving linear subspace detection in high-dimensional spaces demonstrating its capability to effectively model complex geometric structures. Furthermore, the Ebola Optimization Search-driven Adaptive Vision Transformer (EOS-Adaptive VT) uses AVT to learn long-range spatial and contextual dependencies in high-dimensional pattern data, and EOS adaptively explores and exploits the solution space, including 3D geometric registration under rigid transformations and image correspondence estimation. Experimental results shows that the proposed framework consistently performs with 97.24% accuracy, 97.02% precision, and 96.84% recall using Python 3.11. The findings highlight the effectiveness of transformer-based architectures for scalable pattern recognition, particularly in handling complex and high-dimensional datasets. Overall, this research shows the advancement of efficient, scalable, and high-performance pattern recognition systems with applications in computer vision, robotics, and high-dimensional scientific data analysis.