Cognitive Science Fall 2007, Second Midterm

NAME __SOLN______

Cognitive Science Fall 2007, Second Midterm

SOLUTIONS

1. Answer the following true/false questions by circling the correct answer (2 pts each):

a ) True/False- Broca's aphasia is primarily associated with difficulty understanding language.

b ) True/False- Some of the symptoms of Parkinson's and Huntington diseases suggest that syntax and semantics are processed in different parts of the brain.

c ) True/False- Autism is sometimes called "a disease of empathy."

d ) True/False- All semantic memories have some emotional associations.

e ) True/False- Children under the age of four almost always fail the "false belief test."

f ) True/False- Mirror neurons fire when a monkey picks up an object and when the monkey observes a person picking up the object, regardless of how the object is picked up.

g) True/False- A baseball player “practicing” by imagining throwing a ball is an example of resonance.

h) True/False- A person watching a movie of a painful operation may experience some pain. This is an example of simulation.

2. Answer the following multiple choice questions by circling the correct answer. Note: more than one answer may be correct (3 pts each).

a) Both naive Bayes and Markov models make the assumption of conditional independence in order to:

1) allow the application of Bayes rule.

2) realistically model linguistic phenonomena

3) avoid using models with an exponential number of parameters.

4) all of the above.

b) According to the Feature Integration Theory:

1) There are two attentive stages: pre-attentive and post-attentive.

2) The neural basis of attention is associated with sensory cortices.

3) For vision, focused attention allows the integration of information across several feature maps for a given color.

4) For vision, focused attention allows the integration of information across feature maps for a given spatial location.

c) The “what” pathway:

1) Is associated with the dorsal stream.

2) Is associated with the temporal lobe.

3) When damaged, one would expect difficulties in perceiving objects.

4) Is found only in the left hemisphere.

d) Emotional processing has been described in terms of:

1) Feelings, which are most strongly associated with cortical processing.

2) Emotions, which are more strongly associated with the autonomic nervous system.

3) Feelings, which are associated with the autonomic nervous system.

4) Rapid and slow processes, where the former is conscious and the latter unconscious.

e). The number game – the size principle in Tenenbaum's concept learning experiments/anaysis states that:
1) the probability given to each datum in a hypothesis is directly proportional to the size of the hypothesis.
2) the probability given to each datum in a hypothesis is inversely proportional to the size of the hypothesis.
3) the probability of a hypothesis is directly proportional to the size of the class of hypotheses.

4) the probability of a hypothesis is inversely proportional to the size of the class of hypotheses.

3. Define/Short answer questions (points are indicated):

a) (4 pts) Name four of Ekman's six universal emotions

anger, disgust, fear, joy, sadness, and surprise.

b) (4 pts) Define the receptive field of a visual neuron.

A receptive field is that portion of the retina, when stimulated, alters the firing pattern of the neuron.

4) (6 pts) According to Prospect Theory, what is the basis of risk aversion for human decision makers?

It reflects the fact that the value function has a concave shape for gains. The concavity means that a sure loss of $800 vs. an expected loss of $850, people will choose the gamble because it has higher value under the pressures of a concave value function. [Further explanation – the value function has different shapes for gains and losses. For gains the function is concave. So the value associated with a risky bet is not as large as the amount of gain. The loss curve is both steeper than the gain curve, and convex. For the first property, a loss of X dollars is more averse than the gain of X dollars is positive. Thus, there is risk seeking for losses / and risk aversion for gains.]

5) (10 pts). (a) Describe a preconditioning experiment. (b) Does the existence of pre-conditioning more strongly support the Stimulus-Stimulus (SS) or the Stimulus-Response (SR) theory? (c) Consider the other theory (i.e., the theory you did not choose in b). Why does pre-conditioning argue against that theory?

(a) CS1=tone CS2=light US=shock.

Phase 1: pair CS2 à CS1

Phase 2: CS1 à US resulting in fear conditioning

Phase 3: Test CS2.

Result –fear conditioning is observed

b) This supports the S-S response theory.

c) By the S-R theory, the stimulus and response are directly paired. No fear response is expected because the stimulus (CS2) and US were never paired.

6) (7 pts). Imagine that right before you walked into this exam, your competitive friend secretly put some epinephrine into your water bottle, which you've been drinking all along without knowing about it. (a) How might that influence your assessment of how you are performing on the test?

(b) Name the theory you are relying on in answering this question.

a). Because your heart is pounding, you are sweating, etc., you might interpret these events as indicating stress/apprehension. As a result, you might assess your progress as poor (imagine the spiral this might create).

b) Extended James-Lange Theory (James-Lange-Schachter-Damasio) or some variant of that.

7) (14 pts) The computer science department's brand new cappuccino machine "understands" spoken requests for coffee. Unfortunately, the machine hisses and so “s” is not audible to the machine (which means plurals cannot be used to determine whether one or two coffees have been ordered). To process the spoken requests it relies on the following word probability tables, which have been learned through experience at the department:

P(one) = .67

P(two) = .33

P(small|one) = .7 P(small|two) = .2

P(large|one) = .3 P(large|two) = .8

P(espresso(s)|small) = .8

P(espresso(s)|large) = .33

P(cappuccino(s)|small) = .2

P(cappuccino(s)|large) = .67

(a) What are the permissible coffee orders?

One small espr. One small cap One large exp. One large cap.

Two small espr. Two small cap Two large exp. Two large cap.

(b) What is the most frequent coffee order, and what is the probability of this order?

“One small esspresso:” (0.67)(0.7)(0.8) [This just beats: “one large cappuccino”]

(c) You order "Two large cappuccino(s)" but unfortunately there are other people talking in the room and the machine only “hears” the last two words you uttered: "large cappuccino(s)". However, the machine is high-tech and is able to infer the first word you spoke using Bayesian inference. How many cappuccinos will it give you? Show your work!

P(Two | large, cappuccino(s)) = P(two, large, cappuccino(s)) / P(large, cappuccino(s))

= P(two, large, cappuccino(s)) / [P(two, large, cap)+P(one, large, cap)]

à .33 * .8 * .67 / (.33 * .8 * .67 + .67 * .3 * .67) = 0.5677

For “one” the values are: . 67 * .3 * .67 / (.33 * .8 * .67 + .67 * .3 * .67) ; which is smaller than above. Therefore, it will give me two large cappuccinos.

8). (14 pts) Consider the following simple maze, where each letter represents a room (a state):

A-C-Y
| |
B-X

Assume that a rat that uses reinforcement learning and always starts in space A, that space Y always has adequate food in it (reward = 1), and that space X has equal probabilities of food (reward = 1), no food (reward=0), or shock (-1). All other spaces have a reward value of 0. Assume further that the learning constant used is 0.5, and that whenever the rat reaches space X or Y it is removed from the maze with an exit value of 0. Finally, assume that to begin with, the rat associates the value 0 with all spaces of the maze. It makes the following three runs through the maze (being removed from the maze at the end of each run), in the order listed below. What value does the rat associate with each of the spaces it moved through after each run:

i) A -> C -> X (shock)

A: 0 B: 0 C: 0 X: -1/2 Y: 0

ii) A -> B -> X (tasty food)

A: 0 B: -1/4 C: 0 X: 1/4 Y: 0

iii) A -> C -> Y (food)

A: 0 B: -1/4 C: 0 X: 1/4 Y: 1/2

b) If the rat makes a very large number random runs through the maze, will the value be higher for X or Y? Why (briefly)?

The value of X might oscillate for a while, but the expected value at X is zero and the expected value of Y is one. As a result, after many trials the value at X will be smaller.
9) (10 pts) Consider the Von Melchner et al. study with re-wired ferrets. Suppose after rewiring the visual system from left V1 (visual cortex) to the left A1 (auditory cortex) and lesioning the left V1, the ferrets behaved as if they could not see from the contralateral (opposite) side. (a) What research question was Von Melcher et al. addressing? (b) What conclusions ought to be drawn given the result stated above? Why?

a) The theory Von Melchner et al were was whether the development of the cellular organization in sensory corticies (auditory) include extrinsic factors (owing to the pattern of activity in the pathway, the featues of the external stimuli, etc) compared to just intrinsic (i.e., not altered by experience) factors.

b) If the animals appeared to be blind, it would imply the auditory cortex did not re-organize to support visual processing. Under these circumstances the data would not allow the rejection of the hypothesis that the development of cortices are strictly intrinsic.

______

please refrain from writing below this line!

1