> Finally, the formula has a simple interpretation. If we think of the histogram as a probability density function, we can be said to be trying to minimize the

> Finally, the formula has a simple interpretation. If we think of the histogram as a probability density function, we can be said to be trying to minimize the KL divergence between the histogram's distribution and the empirical distribution. Can you expand on that? I mean it does look like the KL divergence from N to T, but you're maximizing it so that doesn't really make sense. According to the paper it's the maximum log likelihood of the samples assuming a piecewise constant rate.