[likelihood] small fixes

This commit is contained in:
Jan Benda 2021-01-11 16:55:54 +01:00
parent 9c0bdba5e6
commit f111d20976

View File

@ -476,12 +476,12 @@ The angle $\phi_{mle}$ that maximizes this likelihood is an estimate
of the true orientation of the bar (\figref{mlecodingfig}).
The noisiness of the neuron's responses as quantified by $\sigma_i$
usually is a function of the neuron's mean firing rate $r_i$,
\eqnref{bartuningcurve}: $\sigma_i = \sigma_i(r_i)$. This dependence
has a major impact of the maximum likelihood estimation. Usually, the
stronger the response of a neuron, the higher its firing rate, the
lower the noise. In this case, strong responses will have a stronger
influence on the position of the maximum of the log-likelihood.
usually is a function of the neuron's mean firing rate $r_i$:
$\sigma_i = \sigma_i(r_i)$. This dependence has a major impact on the
maximum likelihood estimation. Usually, the stronger the response of a
neuron, the higher its firing rate, the lower the noise. In this case,
strong responses have a stronger influence on the position of the
maximum of the log-likelihood.
Whether neural systems really implement maximum likelihood estimators
is another question. There are many ways how a stimulus property can