On Quantized Kernel Affine Projection Algorithm: Sparsification via Delayed Vector Quantization

Quantization techniques compress data by reducing large datasets into smaller code vector sets. This study introduces a vector quantization method within the Kernel Affine Projection Algorithm (KAPA) framework, combining the simplicity and online capabilities of the Kernel Least Mean Squares (KLMS) algorithm with improved performance through reduced gradient noise via multiple feature space samples. The resulting Quantized Kernel Affine Projection Algorithm (Quantized-KAPA) outperforms existing sparsification methods. Simulations involving chaotic time-series forecasting and nonlinear communication channel equalization demonstrate its superior performance.