This paper proposes a simple approach for probabilistic classification problem. Using kernels with random width as basis for logistic regression, and applying automatic relevance determination (ARD) technique, conditional density is estimated nether with cross-validation, nor with gradient-based adaptive parameter tuning.
