Skip to main content
. 2015 Oct 14;11(10):e1004464. doi: 10.1371/journal.pcbi.1004464

Fig 9. Expected LogLikelihood (ELL) based estimation is statistically efficient.

Fig 9

Top (A,B): The ELL-based method (blue) compares favorably to the standard MAP estimate (red) when spikes are fully observed (using the same L1 prior). Bottom (C,D): We compare the ELL based method (blue) to the Expectation Maximization (EM) approach, when only 10% of the spikes are observed. We show the results after one (cyan) and two (magenta) EM steps. The EM steps do not improve over the ELL-based method. Parameters: N = 50, T = 1.4 hours. For the Gibbs sampling we used a single sample after a burn-in period of 30 samples, as we used in our EM simulations without the ELL-based initialization (section E S1 Text).