Learning Performance Analysis of Perceptron Model Based on Markov Sampling

This paper investigates the learning performance of the perceptron model based on Markov sampling, which builds upon the traditional framework with independent and identically distributed samples. Initial efforts establish the constraints on the perceptron model’s learning performance with uniformly ergodic Markov chain samples and validates its consistent behavior. Furthermore, this study introduces a ueMC-PM algorithm. Numerical investigations undertaken on benchmark repositories reveal that the perceptron model utilizing ueMC samples yields fewer misclassification rates.