The author presents an adaptive algorithm for learning drifting discrete distributions, overcoming limitations of previous methods by characterizing statistical error using data-dependent bounds.
