Examiner:

Frank Jäkel, Ulla Martens

Assessor:

Ulla Martens, Frank Jäkel

## Vorbereitung

Ich habe mich zu *Probabilistic Modeling of Perception and Cognition* und zu *Cognitive Psychology* prüfen lassen und vor Anmeldung gefragt, wo jeweils die Schwerpunkte liegen:

**PMPC**: probability theory (vor allem Bayes Law), Schwerpunkt auf Signal Detection Theory.

**Vorbereitung**: Bishop und eigene Aufzeichnungen/CoxiLeaks für Probability Theory; Kapitel 6 aus Coombs et al. 1970 und eigene Aufzeichnungen aus PMPC für den Signal Detection Theory Teil.

**CogPsy**: 7 Abschnitte aus den Vorlesungen: Methods, Perception, Attention, Memory, Language, Emotion and Cognitive Control.

**Vorbereitung**: Slides, selber Fragen überlegt, auswendig gelernt.

## Prüfung

**Ablauf** 30 Minuten, die Sprache kann man sich aussuchen und auch zwischendurch wechseln. Ulla Martens hat 20 Minuten geprüft und zu jedem Abschnitt 1-2 Fragen gestellt. Frank Jäkel hat die restlichen 10 min geprüft und nur Bayes Theorem und SDT behandelt.

Beide waren sehr gut drauf und sehr nett. Es war eine entspannte Atmosphäre.

**Note: ** 1.0

## Questions

### CogPsy

*Methods*

**What signals are measured by fMRI and EEG, respectively?**

Slides.

**Based on that, which method would you use to answer which research question, why?**

Slides.

*Perception*

**How do we perceive movement?**

Explain Corollary Discharge Theory, slides.

*Attention*

**What types of visual selective attention are there?**

Slides.

**Do you know an experimental paradigm to test spatial attention?**

First explain overt vs. covert attention then Posner et al. spatial cueing task and results.

*Memory*

**Explain the Level-of-Processing theory.**

Slides.

**How would you design experiment to test this hypotheses?**

Three conditions: one only physical encoding, one phonological encoding and one semantic encoding. Then test how stable the memory/retrieval is.

**Such a design is prone to confounds, how can we reduce the effect of confounds between the conditions?**

Make sure that the encoding only happens on the respective level (obvious). What she wanted to hear: use the same time window of encoding for each condition, e.g., do not use a longer time window for the semantic condition.

*Language*

**Explain Levelt's theory on language processing**

Slides.

**What is the crucial point of this theory? Is there evidence?**

That the two levels of processing (lemma vs. lexeme level) are distinct. Evidence: Tip-of-the-tongue phenomenon, priming of words: sheep primes goat (semantic, lemma level), goat primes boat (phonological, lexeme level), but sheep does not prime boat (levels are distinct).

*Emotion*

**What do we do if we see something dangerous, a wild animal for example?**

Run away. Explain the model by LeDoux: short, by-passing route from the thalamus to the lateral nucleus of the amygdala primes the percetion of dangerous stimuli, etc.

*Cognitive Control*

**When do we need executive functions, which tasks can be used to test cognitive control?**

Tower of London, Tower of Hanoi, Six Element task. Explain each of them and explain why we need executive functions to solve them or what happens if executive functions do not work properly.

### PMPC

**What is Bayes Rule, can derive it analytically? **

Derive Bayes Rule from product rule, use sum rule to simplify the evidence/normalization to a sum of likelihood and prior.

**What is the idea behind Bayes Rule, can you derive i semantically?**

Draw a joint distribution table of RV X and Y. Explain how the conditional probability conditions on only one column in the table and how this is reflected in Bayes' Rule: division of the joint probability p(x,y) by the marginal probability p(y).

**Now Signal Detection Theory: what is d'?**

First explain the main points of SDT: two roles of the observer, underlying distribution of noise and signal, hits, false alarms, how to shift the decision criterion, how to construct a ROC and why, etc. Then explain d' as the index of sensitivity, e.g., the distance between the means of the underlying distributions. If we use a stronger signal, d' increases because the signal distribution is shifted to the right; the ROC is shift upwards/bowed to the upper left corner.

**In which other disciplines is d' used? In which situation do you use a similar formula?**

Z-transformation, t-testing. What he wanted to hear: in engineering d' is defined as the signal to noise ratio.

### Comment

Maybe there were some more questions that I do not remember. Also, during the examination I answered the questions much more detailed than here. But I think that in most cases the detailed answer can be found in the lecture slides or other literature. If not, feel free to email me.