Wednesday, October 12, 2011

Homework #1

The questions on problem formulation are underspecified, so there's no way to answer them.  These same horseshit questions are in the book as well, and they always cause more problems than they are worth.

Let's be precise about the ambiguity.  Loaded coin:  We know nothing about what the states or the actions are, or even how this pertains to search.  The definition of fully observable is "What the agent can sense at any point in time is sufficient to make the optimal decisions."  But there are no decisions to make: you keep flipping a coin until you get bored, and voila you have your estimate: (See here.) Thus the definition of fully observable does not even apply unless we have some sort of (unstated) cost function to optimize.

For the maze problem, he does not state what the states or actions are so we can't know if it is continuous or not.   Is this a robot in the maze? then it's continuous.  Can we assume actions take us to junctions in the maze? then it is discrete

Problem 6 is a trick question.  What do they hope to find out by asking that?

Problem 7 does not have the correct answer listed for first node to expand.  The first node to expand is a1.  Oh, btw, he didn't state what the actions are (again).


  1. I agree that the homework is fuzzy. If the final tests are as fuzzily constructed, the scores for the students are effectively scores relating the student's ability to guess the context the instructor had in mind... not about the material itself.

    Still, I'm finding the book and the course useful, even if I think that the post about the questions being Zen Koans in the forum very appropriate.

  2. Yes it is confusing indeed. I only disagree in the coin problem, becouse the task is to find a probabilistic model, so like a markov chain to predict the most probable next value of the toss

  3. in my case I view the maze as a graph, as in the example of Romania, then each step an agent makes a decision, it basically solves the problem using the portion of the graph it sees, therefore the states will be discrete. You opened my mind with the robot moving along, which of course makes more sense form a real life perspective. Now I am unsure about the correct answer.

  4. You are right about problem 7. First node to expand is the start node. Another example of ambiguity/carelessness is remark on include goal in list of expanded nodes in homework 1 questions. In algorithms in the text books the node containing goal state is never expanded.

    Another frustrating issue is use of vaguely terms "left" and "right" when talking about graph. The example is:


    From perspective of node "A" which is left "B" or "C"?!

    I am disappointed by the quality of homework questions so far.