[likelihood] 1st version of translation
This commit is contained in:
parent
87f52022c9
commit
413ccf22b3
@ -17,7 +17,7 @@ x_n$ originating from the distribution.
|
||||
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
|
||||
\section{Maximum Likelihood}
|
||||
|
||||
Let $p(x|\theta)$ (to be read as ``Probability(density) of $x$ given
|
||||
Let $p(x|\theta)$ (to be read as ``probability(density) of $x$ given
|
||||
$\theta$.'') the probability (density) distribution of $x$ given the
|
||||
parameters $\theta$. This could be the normal distribution
|
||||
\begin{equation}
|
||||
@ -68,118 +68,119 @@ the likelihood (\enterm{log-likelihood}):
|
||||
|
||||
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
|
||||
\subsection{Example: the arithmetic mean}
|
||||
|
||||
Wenn die Me{\ss}daten $x_1, x_2, \ldots x_n$ der Normalverteilung
|
||||
\eqnref{normpdfmean} entstammen, und wir den Mittelwert $\mu=\theta$ als
|
||||
einzigen Parameter der Verteilung betrachten, welcher Wert von
|
||||
$\theta$ maximiert dessen Likelihood?
|
||||
Suppose that the measurements $x_1, x_2, \ldots x_n$ originate from a
|
||||
normal distribution \eqnref{normpdfmean} and we consider the mean
|
||||
$\mu=\theta$ as the only parameter. Which value of $\theta$ maximizes
|
||||
its likelihood?
|
||||
|
||||
\begin{figure}[t]
|
||||
\includegraphics[width=1\textwidth]{mlemean}
|
||||
\titlecaption{\label{mlemeanfig} Maximum Likelihood Sch\"atzung des
|
||||
Mittelwerts.}{Oben: Die Daten zusammen mit drei m\"oglichen
|
||||
Normalverteilungen mit unterschiedlichen Mittelwerten (Pfeile) aus
|
||||
denen die Daten stammen k\"onnten. Unten links: Die Likelihood
|
||||
in Abh\"angigkeit des Mittelwerts als Parameter der
|
||||
Normalverteilungen. Unten rechts: die entsprechende
|
||||
Log-Likelihood. An der Position des Maximums bei $\theta=2$
|
||||
\"andert sich nichts (Pfeil).}
|
||||
\titlecaption{\label{mlemeanfig} Maximum likelihood estimation of
|
||||
the mean.}{Top: The measured data (blue dots) together with three
|
||||
different possible normal distributions with different means
|
||||
(arrows) that could be the origin of the data. Bootom left: the
|
||||
likelihood as a function of $\theta$ i.e. the mean. It is maximal
|
||||
at a value of $\theta = 2$. Bottom right: the
|
||||
log-likelihood. Taking the logarithm does not change the position
|
||||
of the maximum.}
|
||||
\end{figure}
|
||||
|
||||
Die Log-Likelihood \eqnref{loglikelihood} ist
|
||||
The log-likelihood \eqnref{loglikelihood}
|
||||
\begin{eqnarray*}
|
||||
\log {\cal L}(\theta|x_1,x_2, \ldots x_n)
|
||||
& = & \sum_{i=1}^n \log \frac{1}{\sqrt{2\pi \sigma^2}}e^{-\frac{(x_i-\theta)^2}{2\sigma^2}} \\
|
||||
& = & \sum_{i=1}^n - \log \sqrt{2\pi \sigma^2} -\frac{(x_i-\theta)^2}{2\sigma^2} \; .
|
||||
\end{eqnarray*}
|
||||
Der Logarithmus hat die sch\"one Eigenschaft, die Exponentialfunktion
|
||||
der Normalverteilung auszul\"oschen, da der Logarithmus die
|
||||
Umkehrfunktion der Exponentialfunktion ist ($\log(e^x)=x$).
|
||||
|
||||
Zur Bestimmung des Maximums der Log-Likelihood berechnen wir deren Ableitung
|
||||
nach dem Parameter $\theta$ und setzen diese gleich Null:
|
||||
% FIXME do we need parentheses around the normal distribution in line one?
|
||||
Since the logarithm is the inverse function of the exponential
|
||||
($\log(e^x)=x$), taking the logarithm removes the exponential from the
|
||||
normal distribution. To calculate the maximum of the log-likelihood,
|
||||
we need to take the derivative with respect to $\theta$ and set it to
|
||||
zero:
|
||||
\begin{eqnarray*}
|
||||
\frac{\text{d}}{\text{d}\theta} \log {\cal L}(\theta|x_1,x_2, \ldots x_n) & = & \sum_{i=1}^n - \frac{2(x_i-\theta)}{2\sigma^2} \;\; = \;\; 0 \\
|
||||
\Leftrightarrow \quad \sum_{i=1}^n x_i - \sum_{i=1}^n \theta & = & 0 \\
|
||||
\Leftrightarrow \quad n \theta & = & \sum_{i=1}^n x_i \\
|
||||
\Leftrightarrow \quad \theta & = & \frac{1}{n} \sum_{i=1}^n x_i \;\; = \;\; \bar x
|
||||
\end{eqnarray*}
|
||||
Der Maximum-Likelihood-Sch\"atzer ist das arithmetische Mittel $\bar
|
||||
x$ der Daten. D.h. das arithmetische Mittel maximiert die
|
||||
Wahrscheinlichkeit, dass die Daten aus einer Normalverteilung mit
|
||||
diesem Mittelwert gezogen worden sind (\figref{mlemeanfig}).
|
||||
From the above equations it becomes clear that the maximum likelihood
|
||||
estimation is equivalent to the mean of the data. That is, the
|
||||
assuming the mean of the data as $\theta$ maximizes the likelihood
|
||||
that the data originate from a normal distribution with that mean
|
||||
(\figref{mlemeanfig}).
|
||||
|
||||
|
||||
\begin{exercise}{mlemean.m}{mlemean.out}
|
||||
Ziehe $n=50$ normalverteilte Zufallsvariablen mit einem Mittelwert $\ne 0$
|
||||
und einer Standardabweichung $\ne 1$.
|
||||
|
||||
Plotte die Likelihood (aus dem Produkt der Wahrscheinlichkeiten) und
|
||||
die Log-Likelihood (aus der Summe der logarithmierten
|
||||
Wahrscheinlichkeiten) f\"ur den Mittelwert als Parameter. Vergleiche
|
||||
die Position der Maxima mit dem aus den Daten berechneten
|
||||
Mittelwert.
|
||||
Draw $n=50$ random numbers from a normal distribution with a mean of
|
||||
$\ne 0$ and a standard deviation of $\ne 1$.
|
||||
|
||||
Plot the likelihood (the product of the probabilities) and the
|
||||
log-likelihood (given by the sum of the logarithms of the
|
||||
probabilities) for the mean as parameter. Compare the position of
|
||||
the maxima with the mean calculated from the data.
|
||||
\pagebreak[4]
|
||||
\end{exercise}
|
||||
|
||||
|
||||
\pagebreak[4]
|
||||
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
|
||||
\section{Kurvenfit als Maximum-Likelihood Sch\"atzung}
|
||||
Beim \determ{Kurvenfit} soll eine Funktion $f(x;\theta)$ mit den Parametern
|
||||
$\theta$ an die Datenpaare $(x_i|y_i)$ durch Anpassung der Parameter
|
||||
$\theta$ gefittet werden. Wenn wir annehmen, dass die $y_i$ um die
|
||||
entsprechenden Funktionswerte $f(x_i;\theta)$ mit einer
|
||||
Standardabweichung $\sigma_i$ normalverteilt streuen, dann lautet die
|
||||
Log-Likelihood
|
||||
\section{Curve fitting as using maximum-likelihood estimation}
|
||||
|
||||
During curve fitting a function of the form $f(x;\theta)$ with the
|
||||
parameter $\theta$ is adapted to the data pairs $(x_i|y_i)$ by
|
||||
adapting $\theta$. When we assume that the $y_i$ values are normally
|
||||
distributed around the function values $f(x_i;\theta)$ with a standard
|
||||
deviation $\sigma_i$, the log-likelihood is
|
||||
|
||||
\begin{eqnarray*}
|
||||
\log {\cal L}(\theta|(x_1,y_1,\sigma_1), \ldots, (x_n,y_n,\sigma_n))
|
||||
& = & \sum_{i=1}^n \log \frac{1}{\sqrt{2\pi \sigma_i^2}}e^{-\frac{(y_i-f(x_i;\theta))^2}{2\sigma_i^2}} \\
|
||||
& = & \sum_{i=1}^n - \log \sqrt{2\pi \sigma_i^2} -\frac{(y_i-f(x_i;\theta))^2}{2\sigma_i^2} \\
|
||||
\end{eqnarray*}
|
||||
Der einzige Unterschied zum vorherigen Beispiel ist, dass die
|
||||
Mittelwerte der Normalverteilungen nun durch die Funktionswerte
|
||||
gegeben sind.
|
||||
|
||||
Der Parameter $\theta$ soll so gew\"ahlt werden, dass die
|
||||
Log-Likelihood maximal wird. Der erste Term der Summe ist
|
||||
unabh\"angig von $\theta$ und kann deshalb bei der Suche nach dem
|
||||
Maximum weggelassen werden:
|
||||
The only difference to the previous example is that the averages in
|
||||
the equations above are now given as the function values
|
||||
$f(x_i;\theta)$
|
||||
|
||||
The parameter $\theta$ should be the one that maximizes the
|
||||
log-likelihood. The first part of the sum is independent of $\theta$
|
||||
and can thus be ignored during the search of the maximum:
|
||||
\begin{eqnarray*}
|
||||
& = & - \frac{1}{2} \sum_{i=1}^n \left( \frac{y_i-f(x_i;\theta)}{\sigma_i} \right)^2
|
||||
\end{eqnarray*}
|
||||
Anstatt nach dem Maximum zu suchen, k\"onnen wir auch das Vorzeichen der Log-Likelihood
|
||||
umdrehen und nach dem Minimum suchen. Dabei k\"onnen wir auch den Faktor $1/2$ vor der Summe vernachl\"assigen --- auch das \"andert nichts an der Position des Minimums:
|
||||
We can further simplify by inverting the sign and then search for the
|
||||
minimum. Also the $1/2$ factor can be ignored since it does not affect
|
||||
the position of the minimum:
|
||||
\begin{equation}
|
||||
\label{chisqmin}
|
||||
\theta_{mle} = \text{argmin}_{\theta} \; \sum_{i=1}^n \left( \frac{y_i-f(x_i;\theta)}{\sigma_i} \right)^2 \;\; = \;\; \text{argmin}_{\theta} \; \chi^2
|
||||
\end{equation}
|
||||
Die Summe der quadratischen Abst\"ande normiert auf die jeweiligen
|
||||
Standardabweichungen wird auch mit $\chi^2$ bezeichnet. Der Wert des
|
||||
Parameters $\theta$, welcher den quadratischen Abstand minimiert, ist
|
||||
also identisch mit der Maximierung der Wahrscheinlichkeit, dass die
|
||||
Daten tats\"achlich aus der Funktion stammen k\"onnen. Minimierung des
|
||||
$\chi^2$ ist also eine Maximum-Likelihood Sch\"atzung.
|
||||
|
||||
An der Herleitung sehen wir aber auch, dass die Minimierung des
|
||||
quadratischen Abstands nur dann eine Maximum-Likelihood Absch\"atzung
|
||||
ist, wenn die Daten normalverteilt um die Funktion streuen. Bei
|
||||
anderen Verteilungen m\"usste man die Log-Likelihood entsprechend
|
||||
\eqnref{loglikelihood} ausrechnen und maximieren.
|
||||
The sum of the squared differences when normalized to the standard
|
||||
deviation is also called $\chi^2$. The parameter $\theta$ which
|
||||
minimizes the squared differences is thus the one that maximizes the
|
||||
probability that the data actually originate from the given
|
||||
function. Minimizing $\chi^2$ therefore is a maximum likelihood
|
||||
estimation.
|
||||
|
||||
From the mathematical considerations above we can see that the
|
||||
minimization of the squared difference is a maximum-likelihood
|
||||
estimation only if the data are normally distributed around the
|
||||
function. In case of other distributions, the log-likelihood needs to
|
||||
be adapted accordingly \eqnref{loglikelihood} and be maximized
|
||||
respectively.
|
||||
|
||||
\begin{figure}[t]
|
||||
\includegraphics[width=1\textwidth]{mlepropline}
|
||||
\titlecaption{\label{mleproplinefig} Maximum-Likelihood Sch\"atzung der
|
||||
Steigung einer Ursprungsgeraden.}{}
|
||||
\titlecaption{\label{mleproplinefig} Maximum likelihood estimation
|
||||
of the slope of line through the origin.}{}
|
||||
\end{figure}
|
||||
|
||||
|
||||
\subsection{Beispiel: einfache Proportionalit\"at}
|
||||
Als Funktion nehmen wir die Ursprungsgerade
|
||||
\subsection{Example: simple proportionality}
|
||||
The function of a line going through the origin
|
||||
\[ f(x) = \theta x \]
|
||||
mit Steigung $\theta$. Die $\chi^2$-Summe lautet damit
|
||||
with the slope $\theta$. The $\chi^2$-sum is thus
|
||||
\[ \chi^2 = \sum_{i=1}^n \left( \frac{y_i-\theta x_i}{\sigma_i} \right)^2 \; . \]
|
||||
Zur Bestimmung des Minimums berechnen wir wieder die erste Ableitung nach $\theta$
|
||||
und setzen diese gleich Null:
|
||||
To estimate the minimum we again take the first derivative with
|
||||
respect to $\theta$ and equate it to zero:
|
||||
\begin{eqnarray}
|
||||
\frac{\text{d}}{\text{d}\theta}\chi^2 & = & \frac{\text{d}}{\text{d}\theta} \sum_{i=1}^n \left( \frac{y_i-\theta x_i}{\sigma_i} \right)^2 \nonumber \\
|
||||
& = & \sum_{i=1}^n \frac{\text{d}}{\text{d}\theta} \left( \frac{y_i-\theta x_i}{\sigma_i} \right)^2 \nonumber \\
|
||||
@ -188,123 +189,124 @@ und setzen diese gleich Null:
|
||||
\Leftrightarrow \quad \theta \sum_{i=1}^n \frac{x_i^2}{\sigma_i^2} & = & \sum_{i=1}^n \frac{x_iy_i}{\sigma_i^2} \nonumber \\
|
||||
\Leftrightarrow \quad \theta & = & \frac{\sum_{i=1}^n \frac{x_iy_i}{\sigma_i^2}}{ \sum_{i=1}^n \frac{x_i^2}{\sigma_i^2}} \label{mleslope}
|
||||
\end{eqnarray}
|
||||
Damit haben wir nun einen anlytischen Ausdruck f\"ur die Bestimmung
|
||||
der Steigung $\theta$ der Regressionsgeraden gewonnen
|
||||
(\figref{mleproplinefig}).
|
||||
|
||||
Ein Gradientenabstieg ist f\"ur das Fitten der Geradensteigung also
|
||||
gar nicht n\"otig. Das gilt allgemein f\"ur das Fitten von
|
||||
Koeffizienten von linear kombinierten Basisfunktionen. Wie z.B.
|
||||
die Steigung $m$ und der y-Achsenabschnitt $b$ einer Geradengleichung
|
||||
With this we obtained an analytical expression for the estimation of
|
||||
the slope $\theta$ of the regression line (\figref{mleproplinefig}).
|
||||
|
||||
A gradient descent, as we have done in the previous chapter, is thus
|
||||
not necessary for the fitting of the slope of a linear equation. This
|
||||
holds even more generally for fitting the coefficients of linearly
|
||||
combined basis functions as for example the fitting of the slope $m$
|
||||
and the y-intercept $b$ of the linear equation
|
||||
\[ y = m \cdot x +b \]
|
||||
oder allgemeiner die Koeffizienten $a_k$ eines Polynoms
|
||||
or, more generally, the coefficients $a_k$ of a polynom
|
||||
\[ y = \sum_{k=0}^N a_k x^k = a_o + a_1x + a_2x^2 + a_3x^4 + \ldots \]
|
||||
\matlabfun{polyfit()}.
|
||||
|
||||
Parameter, die nichtlinear in einer Funktion enthalten sind, k\"onnen
|
||||
im Gegensatz dazu nicht analytisch aus den Daten berechnet
|
||||
werden. z.B. die Rate $\lambda$ eines exponentiellen Zerfalls
|
||||
Parameters that are non-linearly combined can not be calculated
|
||||
analytically. Consider for example the rate $\lambda$ of the
|
||||
exponential decay
|
||||
\[ y = c \cdot e^{\lambda x} \quad , \quad c, \lambda \in \reZ \; . \]
|
||||
F\"ur diesen Fall bleibt dann nur auf numerische Verfahren zur
|
||||
Optimierung der Kostenfunktion, wie z.B. der Gradientenabstieg,
|
||||
zur\"uckzugreifen \matlabfun{lsqcurvefit()}.
|
||||
Such cases require numerical solutions for the optimization of the
|
||||
cost function, e.g. the gradient descent \matlabfun{lsqcurvefit()}.
|
||||
|
||||
|
||||
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
|
||||
\section{Fits von Wahrscheinlichkeitsverteilungen}
|
||||
Jetzt betrachten wir noch den Fall, bei dem wir die Parameter einer
|
||||
Wahrscheinlichkeitsdichtefunktion (z.B. den shape-Parameter einer
|
||||
\determ{Gamma-Verteilung}) an ein Datenset fitten wollen.
|
||||
|
||||
Ein erster Gedanke k\"onnte sein, die
|
||||
\determ[Wahrscheinlichkeitsdichte]{Wahrscheinlichkeitsdichtefunktion}
|
||||
durch Minimierung des quadratischen Abstands an ein Histogramm der
|
||||
Daten zu fitten. Das ist aber aus folgenden Gr\"unden nicht die
|
||||
Methode der Wahl: (i) Wahrscheinlichkeitsdichten k\"onnen nur positiv
|
||||
sein. Darum k\"onnen insbesondere bei kleinen Werten die Daten nicht
|
||||
symmetrisch streuen, wie es bei normalverteilten Daten der Fall
|
||||
ist. (ii) Die Datenwerte sind nicht unabh\"angig, da das normierte
|
||||
Histogram sich zu Eins aufintegriert. Die beiden Annahmen
|
||||
normalverteilte und unabh\"angige Daten, die die Minimierung des
|
||||
quadratischen Abstands \eqnref{chisqmin} zu einem Maximum-Likelihood
|
||||
Sch\"atzer machen, sind also verletzt. (iii) Das Histogramm h\"angt
|
||||
von der Wahl der Klassenbreite ab (\figref{mlepdffig}).
|
||||
Finally let's consider the case in which we want to fit the parameters
|
||||
of a probability density function (e.g. the shape parameter of a
|
||||
\enterm{Gamma-distribution}) to a dataset.
|
||||
|
||||
A first guess could be to fit the probability density by minimization
|
||||
of the squared difference to a histogram of the measured data. For
|
||||
several reasons this is, however, not the method of choice: (i)
|
||||
probability densities can only be positive which leads, for small
|
||||
values in particular, to asymmetric distributions. (ii) the values of
|
||||
a histogram are not independent because the integral of a density is
|
||||
unity. The two basic assumptions of normally distributed and
|
||||
independent samples, which are a prerequisite make the minimization of
|
||||
the squared difference \eqnref{chisqmin} to a maximum likelihood
|
||||
estimation, are violated. (iii) The histogram strongly depends on the
|
||||
chosen bin size \figref{mlepdffig}).
|
||||
|
||||
\begin{figure}[t]
|
||||
\includegraphics[width=1\textwidth]{mlepdf}
|
||||
\titlecaption{\label{mlepdffig} Maximum-Likelihood Sch\"atzung einer
|
||||
Wahrscheinlichkeitsdichtefunktion.}{Links: die 100 Datenpunkte, die
|
||||
aus der Gammaverteilung 2. Ordnung (rot) gezogen worden sind. Der
|
||||
Maximum-Likelihood-Fit ist orange dargestellt. Rechts: das
|
||||
normierte Histogramm der Daten zusammen mit dem \"uber Minimierung
|
||||
des quadratischen Abstands zum Histogramm berechneten Fit.}
|
||||
\titlecaption{\label{mlepdffig} Maximum likelihood estimation of a
|
||||
probability density.}{Left: the 100 data points drawn from a 2nd
|
||||
order Gamma-distribution. The maximum likelihood estimation of the
|
||||
probability density function is shown in orange, the true pdf is
|
||||
shown in red. Right: normalized histogram of the data together
|
||||
with the real (red) and the fitted probability density
|
||||
functions. The fit was done by minimizing the squared difference
|
||||
to the histogram.}
|
||||
\end{figure}
|
||||
|
||||
Den direkten Weg, eine Wahrscheinlichkeitsdichtefunktion an ein
|
||||
Datenset zu fitten, haben wir oben schon bei dem Beispiel zur
|
||||
Absch\"atzung des Mittelwertes einer Normalverteilung gesehen ---
|
||||
Maximum Likelihood! Wir suchen einfach die Parameter $\theta$ der
|
||||
gesuchten Wahrscheinlichkeitsdichtefunktion bei der die Log-Likelihood
|
||||
\eqnref{loglikelihood} maximal wird. Das ist im allgemeinen ein
|
||||
nichtlinieares Optimierungsproblem, das mit numerischen Verfahren, wie
|
||||
z.B. dem Gradientenabstieg, gel\"ost wird \matlabfun{mle()}.
|
||||
|
||||
Using the example of the estimating the mean value of a normal
|
||||
distribution we have discussed the direct approach to fit a
|
||||
probability density to data via maximum likelihood. We simply search
|
||||
for the parameter $\theta$ of the desired probability density function
|
||||
that maximizes the log-likelihood. This is a non-linear optimization
|
||||
problem that is generally solved with numerical methods such as the
|
||||
gradient descent \matlabfun{mle()}.
|
||||
|
||||
\begin{exercise}{mlegammafit.m}{mlegammafit.out}
|
||||
Erzeuge Gammaverteilte Zufallszahlen und benutze Maximum-Likelihood,
|
||||
um die Parameter der Gammafunktion aus den Daten zu bestimmen.
|
||||
Create a sample of gamma-distributed random number and apply the
|
||||
maximum likelihood method to estimate the parameters of the gamma
|
||||
function from the data.
|
||||
\pagebreak
|
||||
\end{exercise}
|
||||
|
||||
|
||||
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
|
||||
\section{Neuronale Kodierung}
|
||||
In sensorischen Systemen kodieren Populationen von Neuronen mit ihrer
|
||||
Aktivit\"at Eigenschaften von sensorischen Stimuli. z.B. im visuellen
|
||||
Kortex V1 die Orientierung eines Balkens. Traditionell wird die
|
||||
Antwort der Neurone f\"ur verschiedene Stimuli (z.B. verschiedene
|
||||
Orientierungen des Balkens) gemessen. Die mittlere Antwort der Neurone
|
||||
als Funktion eines Stimulusparameters ist dann die
|
||||
\enterm{Tuning-curve} (deutsch \determ{Abstimmkurve}, z.B. Feuerrate
|
||||
als Funktion des Orientierungswinkels).
|
||||
\section{Neural coding}
|
||||
In sensory systems certain aspects of the surrounding are encoded in
|
||||
the neuronal activity of populations of neurons. One example of such
|
||||
population coding is the tuning of neurons in the primary visual
|
||||
cortex (V1) to the orientation of a visual stimulus. Different neurons
|
||||
respond best to different stimulus orientations. Traditionally, such a
|
||||
tuning is measured by analyzing the neuronal response strength
|
||||
(e.g. the firing rate) as a function of the orientation of the visual
|
||||
stimulus and is depicted and summarized with the so called
|
||||
\enterm{tuning-curve}(German \determ{Abstimmkurve},
|
||||
figure~\ref{mlecoding}, top).
|
||||
|
||||
\begin{figure}[tp]
|
||||
\includegraphics[width=1\textwidth]{mlecoding}
|
||||
\titlecaption{\label{mlecodingfig} Maximum Likelihood Sch\"atzung
|
||||
eines Stimulusparameters aus neuronaler Aktivit\"at.}{Oben:
|
||||
Die Tuning-Kurve eines einzelnen Neurons in Abh\"angigkeit von der
|
||||
Orientierung eines Balkens. Der Stimulus der die st\"akste
|
||||
Aktivit\"at in diesem Neuron hervorruft ist ein senkrechter Balken
|
||||
(Pfeil, $\phi_i=90$\,\degree. Die rote Fl\"ache deutet die
|
||||
Variabilit\"at $p(r)$ der Aktivit\"at $r$ um die Tuning-Kurve
|
||||
herum an. Mitte: Jedes Neuron in der Population hat eine andere
|
||||
bevorzugte Orientierung des Stimulus (farbige Linien). Ein
|
||||
Stimulus einer bestimmten Orientierung aktiviert die Neurone in
|
||||
spezifischer Weise (Punkte). Unten: Die Log-Likelihood dieser
|
||||
Aktivit\"aten wird in der N\"ahe der wahren Orientierung
|
||||
des Stimulus maximiert.}
|
||||
\titlecaption{\label{mlecodingfig} Maximum likelihood estimation of
|
||||
a stimulus parameter from neuronal activity.}{Top: Tuning curve of
|
||||
an individual neuron as a function of the stimulus orientation (a
|
||||
dark bar in front of a white background). The stimulus that evokes
|
||||
the strongest activity in that neuron is the bar with the vertical
|
||||
orientation (arrow, $\phi_i=90$\,\degree). The red area indicates
|
||||
the variability of the neuronal activity $p(r)$ around the tunig
|
||||
curve. Center: In a population of neurons, each neuron may have a
|
||||
different tuning curve (colors). A specific stimulus (the vertical
|
||||
bar) activates the individual neurons of the population in a
|
||||
specific way (dots). Bottom: The log-likelihood of the activity
|
||||
pattern will be maximized close to the real stimulus orientation.}
|
||||
\end{figure}
|
||||
|
||||
Das Gehirn ist aber mit dem umgekehrten Problem konfrontiert: gegeben
|
||||
eine bestimmte Aktivit\"at der Neurone in der Population, was war der
|
||||
Stimulus (die Orientierung des Balkens)? Eine m\"ogliche Antwort ist
|
||||
im Sinne von Maximum-Likelihood: es war der Stimulus f\"ur den das
|
||||
Aktivit\"atsmuster am wahrscheinlichsten ist.
|
||||
|
||||
Bleiben wir mit einem Beispiel bei den orientierungssensitiven Zellen
|
||||
des V1. Das Tuning $\Omega_i(\phi)$ der Zellen $i$ auf ihre bevorzugte
|
||||
Orientierung $\phi_i$ l\"asst sich gut mit einer van-Mises Funktion
|
||||
(entspricht der Gaussfunktion auf einer zyklischen x-Achse)
|
||||
beschreiben (\figref{mlecodingfig}):
|
||||
The brain, however, is confronted with the inverse problem: given a
|
||||
certain activity pattern in the neuronal population, what was the
|
||||
stimulus? In the sense of maximum likelihood, a possible answer to
|
||||
this question would be: It was the stimulus for which the particular
|
||||
activity pattern is most likely.
|
||||
|
||||
Let's stay with the example of the orientation tuning in V1. The
|
||||
tuning $\Omega_i(\phi)$ of the neurons $i$ to the preferred stimulus
|
||||
orientation $\phi_i$ can be well described using a van-Mises function
|
||||
(the Gaussian function on a cyclic x-axis) (\figref{mlecodingfig}):
|
||||
\[ \Omega_i(\phi) = c \cdot e^{\cos(2(\phi-\phi_i))} \quad , \quad c
|
||||
\in \reZ \]
|
||||
Die Aktivit\"at der Neurone approximieren wir hier mit einer
|
||||
Normalverteilung um die Tuning-Kurve mit Standardabweichung
|
||||
$\sigma=\Omega/4$ proportional zu $\Omega$, so dass die
|
||||
Wahrscheinlichkeit $p_i(r|\phi)$ des $i$-ten Neurons die Aktivit\"at $r$ zu
|
||||
haben, wenn ein Stimulus mit Orientierung $\phi$ anliegt, gegeben ist durch
|
||||
\in \reZ \]
|
||||
The neuronal activity is approximated with a normal
|
||||
distribution around the tuning curve with a standard deviation
|
||||
$\sigma=\Omega/4$ which is proprotional to $\Omega$ such that the
|
||||
probability $p_i(r|\phi)$ of the $i$-th neuron showing the activity
|
||||
$r$ given a certain orientation $\phi$ is given by
|
||||
|
||||
\[ p_i(r|\phi) = \frac{1}{\sqrt{2\pi}\Omega_i(\phi)/4} e^{-\frac{1}{2}\left(\frac{r-\Omega_i(\phi)}{\Omega_i(\phi)/4}\right)^2} \; . \]
|
||||
Die Log-Likelihood der Stimulusorientierung $\phi$ gegeben die
|
||||
Aktivit\"aten $r_1$, $r_2$, ... $r_n$ ist damit
|
||||
The log-likelihood of the stimulus orientation $\phi$ given the
|
||||
activity pattern in the population $r_1$, $r_2$, ... $r_n$ is thus
|
||||
\[ {\cal L}(\phi|r_1, r_2, \ldots r_n) = \sum_{i=1}^n \log p_i(r_i|\phi) \]
|
||||
|
||||
\selectlanguage{english}
|
||||
|
Reference in New Issue
Block a user