From Adaptive Kernel Density Estimation to Sparse Mixture Models