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

Automated excerpt

The KL divergence corresponds to \(f(t) = t \log t \, – t + 1\) and \(f^\ast(u) = e^u \, – 1\). Top: using linear features, bottom: using quadratic features. From left to right: softmax regression, spectral estimation for KL divergence, spectral estimation for Pearson divergence.

Selected automatically from source text; not independently written or fact-checked. Read the original for full context.