Exploding variance of means of exponentials: least-squares to the rescue

Francis Bach explores the exploding variance of means of exponentials in the context of spectral log density estimation, and introduces a new algorithm that uses least-squares to mitigate this issue. The algorithm is based on a combination of spectral and temporal information, and is particularly useful for applications where the log density of the underlying process is smooth and can be well-approximated by a low-order polynomial.