====================================================================

CS/Psych/ISyE 6795
Cognitive Scvience
Spring 2005
Ashok K. Goel

Mid-Term Examination
Assigned: Wednesday, March 9, 2005
Due: Monday, March 14, 2005 (by class-time)

--------------------------------------------------------------------

General Notes: You may consult any textbook you wish including the two
textbooks for the class. You may also consult your classnotes
including the classnotes posted on the class webpage/swicki. Of course,
you want to cite the references you use.

However, discussion with any other person is strictly off-limits.

Please make your answer specific and precise, and illustrate them with
diagrams and examples:  precision and clarity are critical.

Please answer all questions.

====================================================================

Question 1 (20 points):

A Scandal in Bohemia is often is regarded as
one of the best Sherlock Holmes stories written by Sir Arthur Conan
Doyle. The story can be accessed from the pull-down menu at the
Sherlock Holmes Museum at: http://www.sherlock-holmes.co.uk/home.htm
(Many other websites also contain links to this story.)

1(a): What are deductive, abductive, inductive and predictive
inferences?

1(b) Analyze Scandal in Bohemia in terms of deduction, abduction,
induction, and prediction, i.e., identify the segments in which Holmes
(or one of the other characters) is engaging in deductive, abductive,
inductive, or predictive inferences.

1(c): What, if anything, does your analyis of this story imply for
human cognition? You may want to consider various perspectives on
this, e.g., the perspectives of the main characters and also the
perspectives of the author and the reader (i.e., you).

====================================================================

Question 2 (10 points):

Any knowledge that is represented as logical assertions can be
converted to a probabilistic representation simply by assigning
probability 1 to all True assertions and probability 0 to all False
assertions. The reverse is not true however: hypotheses with
probabilities between 0 and 1 are not directly representable as
logical assertions.

If probabilities can represent everything that logic can, and more,
why build a theory of cognition that uses logic as a foundation for
knowledge representation?  What arguments, if any, may support the
use of logic over probability? Conversely, what arguments, if any, may
support the use of probability over logic?

====================================================================

Question 3 (10 points):

In class, we discussed two meanings of concepts: concepts as
categories and concepts as models.

3(a): Do concepts viewed as models also act like categories? If so,
what is the difference between concepts as categories and concepts as
models?

3(b): Categorization appears to be ubiquituous in cognition. Why? That is,
what is the power of categorization?

====================================================================

Question 4 (10 points):

Both case-based and analogical reasoning involve reminding and old
problems and transfer of parts of their solutions to new
problems. What (if any) are the differences between the two? For what
class of problems or situations is case-based reasoning most
effective? Analogical Reasoning?

====================================================================

Question 5: (15 points):

In class, we discussed a multilayered neural network with inhibitory
and excitory connections for word recognition.  This network
recognized words even if the input was incomplete or partially
incorrect. For example, if presented with the input of BEFR with the
intended input of BEER, a network capable of recognizing the words
BEER and BEEP may correctly recognize the input to be BEER.

3(a) What changes (if any) to the above architecture would be needed to
recognize partially incorrect sentences instead of words, e.g., given
the the input of "I WOULD LIKE A COLD BEEP" with the intended input of
I WOULD LIKE A COLD BEER."

3(b) Draw a schematic of the new architecture.

3(c) Is this method cognitively plausible for this problem. Why (or
why not)?  Please explain.

====================================================================

Question 6 (10 points):

Marr's theory of visual perception decomposes the task of object
recognition in vision into several subtasks.  What is this task
decomposition?  What motivates the decomposition? Why not simply use
the "bitmap" like raw image as a probe into the long-term memory of
images and their interpretations, retrieve the closest matching image,
and decide on an interpretation of the current image?

====================================================================

Question 7 (15 points):

A main issue in the imagery debate is whether mental representations
of spatial knowledge are descriptive or depictive. Recently there has
been quite a bit of neurobiological evidence in favor of depictive
representations.

4(a) What is the neurobilogical evidence in favor of depictive representations?

4(b) Does this evidence resolve the debate? Why (or why not)?

====================================================================

Question 8 (5 points):

It is often argued that English (or any other natural language) cannot be
the Language of Thought. Why not?

====================================================================