Kernel Density Estimation
The most classical nonparametric density estimation method, which places a kernel function, typically a symmetric probability density such as a Gaussian kernel, at each data point and sums and normalizes all kernel functions to obtain a density estimate. The convergence rate is determined by the bandwidth parameter, attaining the optimal rate of n^{-s/(2s+1)} for densities with smoothness s.
Timeline (last 90 days)
Estimation in mode regression is usually implemented via kernel density estimation or empirical likelihood methods, and its statistical properties are more complex than those of mean regression, with the convergence rate depending on the curvature of the density near the peak.
The paper proves that a continuous plug-in estimator based on kernel density estimation achieves the minimax rate n^{-s/(d+2s)} when the support manifolds are known.
Kernel density estimation achieves a convergence rate of n^{-s/(2s+1)} on densities with smoothness s.
In kernel density estimation (KDE), when the Gaussian kernel bandwidth parameter σ is too small the estimate degenerates into a collection of discrete points, while when σ is too large it over-smooths and loses detailed information, reflecting the bias-variance tradeoff.
Kernel Density Estimation (KDE) degrades sharply when dimensions exceed 5–10, because the required sample size grows exponentially with dimensionality.
All Facts (5)
Estimation in mode regression is usually implemented via kernel density estimation or empirical likelihood methods, and its statistical properties are more complex than those of mean regression, with the convergence rate depending on the curvature of the density near the peak.
50%UnverifiedThe paper proves that a continuous plug-in estimator based on kernel density estimation achieves the minimax rate n^{-s/(d+2s)} when the support manifolds are known.
50%UnverifiedKernel density estimation achieves a convergence rate of n^{-s/(2s+1)} on densities with smoothness s.
50%UnverifiedIn kernel density estimation (KDE), when the Gaussian kernel bandwidth parameter σ is too small the estimate degenerates into a collection of discrete points, while when σ is too large it over-smooths and loses detailed information, reflecting the bias-variance tradeoff.
50%UnverifiedKernel Density Estimation (KDE) degrades sharply when dimensions exceed 5–10, because the required sample size grows exponentially with dimensionality.
50%