\documentclass[a4paper,12pt,pdftex]{exam}
\newcommand{\ptitle}{Face-selectivity index}
\input{../header.tex}
\firstpagefooter{Supervisor: Marius G\"orner}{}%
{email: marius.goerner@uni-tuebingen.de}
\begin{document}
\input{../instructions.tex}
%%%%%%%%%%%%%% Questions %%%%%%%%%%%%%%%%%%%%%%%%%
\section{Estimating the selectivity index (SI) of neurons}
In the temporal lobe of primates you can find neurons that respond
selectively to a certain type of object category. You may have heard
about the famous grandmother neuron which is supposed to respond
exclusively when the subject sees a particular person, i.e. the
grandmother. Even though the existence of a grandmother neuron in the
strict sense is implausible, the concept exemplifies the observation
that sensory neurons within the ventral visual stream are tuned to
certain stimulus types. One of the most important and first visual
stimulus the newborn typically sees is the mother's face. It is
believed that the early ubiquity of faces and their importance for
social interactions triggers the development of the so called
face-patch system within the temporal lobe of primates.\par Your task
here will be to estimate the \textit{selectivity index} ($SI$)
of neurons that were recorded in the superior temporal sulcus of a
rhesus monkey during the visual presentation of objects of different
categories (data courtesy of the Sensorymotor-Lab, Hertie
Institute).
\begin{questions}
\question
In the accompanying datasets you find the
\texttt{spiketimes} of 184 neurons that were recorded during the visual
presentation of non-face like stimuli (tools, fruits, hands and
bodies) and averted and directed faces of humans and rhesus
monkeys. Each \texttt{.mat}-file contains the data of one neuron
which was recorded during multiple trials. Spike times are given in
ms relative to trial onset. The field \texttt{spiketimes} contains
cells that contain the the data of the trails. Each trial consists of
400 ms of baseline recording (presentation of white noise) followed by 400 ms
of stimulus presentation. Each trial belongs to one object category,
trial identities can be found in the \texttt{*\_trials}-fields
(9 fields).
\begin{parts}
\part
Illustrate the spiking activity of 10 neurons, sorted by object
category, in a raster plot. As a result you should get one plot
for each neuron subdevided in subplots for the different
categories. Mind that there are four categories that contain faces
(\texttt{averted\_human}, \texttt{face} (straight human face),
\texttt{monkey} (straight monkey face) and \texttt{gaze\_monkey}),
you may want to analyze them separately as well as combined. Add
also a marker where the stimulus starts.
\part
Estimate the time-resolved firing rate of each neuron for each
object category. Use at least two different methods
(e.g. instantaneous firing rate based on interspike intervals,
spike counting within bins (PSTH), kernel density estimation). Do
this individually for each trial and average afterwards in order
to obtain the standard deviation of the firing rates. For the 10
neurons that you plotted above plot the firing rates and their
standard deviations on top of the raster plots. Which of the
methods appears best to represent the spiking activity seen in the
raster plots?
\part
Generate a figure that shows for 20 neurons the firing rates
belonging to each object category. Don't forget to add an
appropriate legend.
\part
Next step is to examine the obtained firing rates for significant
modulations.
% First, normalize each response to baseline activity
% (first 400 ms). Why is the normalization useful?
Determine the periods within which the neurons activity deviates
from the baseline activity at least by $2*\sigma$. Do this for
each object category and mark the periods in the plots in an
appropriate way. Are there neurons that do not respond to the
visual stimulation or exhibit inhibitory responses?
\par
\part
The $SI$ gives an estimate of how strongly a neuron is tuned to
the chosen object categories. It is given by the neuron's response
during the presentation of one category compared to another
category.
\begin{equation}
SI = \frac{ \mu_{\text{Response to category A}} - \mu_{ \text{Response
to category B}} } { \mu_{\text{Response to category A}} + \mu_{ \text{Response
to category B} } }
\end{equation}
$SI$ can take values between -1 and 1 which indicates tuning to
the one or to the other category. There are different
possibilities of how it can be estimated. The easiest way would be
to average the spike count during the whole time of stimulus
presentation. However, if responses are phasic you will
underestimate the $SI$. Therefore, you should limit the estimate
to periods of significant modulations. Use the periods determined
in (d). Why may using the value of the peak activity be inappropriate?
Store all obtained $SI$s within a single variable. We are mainly
interested in identifying face-selective neurons but feel free to
test the neurons for selectivity to other categories, as well.
\part
Plot the distribution of $SI$ values and describe it
qualitatively. Does it indicate a continuum or a distinct
population of face-selective neurons. \par Think about a
statistical test that tells you whether a given neuron is
significantly modulated by one or the other category (try
different combinations of categories). List cells that show
significant modulations to faces and non-faces. Which is the
minimum $SI$ that reaches significance when choosing
$\alpha = 0.05$? Is it an all or nothing selectivity?
\part
Take a look at the time resolved firing rates of the identified
face-selective neurons and examine their response properties. What
are their response-latencies (choose an appropriate
visualisation), are their responses phasic or tonic.
\end{parts}
\end{questions}
\end{document}