Saturday, October 15, 2011

Artificial Intelligence Class Errata

I've run across a few minor errata in the lecture material and homework questions for the Stanford Artificial Intelligence class so far. Pretty impressive quality considering the volume and sophistication of the content. I'll save the homework question errata until after the answers are released, but here are some of the others:

Unit 02-16 01:08/01:14 -- Uniform Cost Search 4
Total cost is computed by Peter Norvig as 460 when it is really 450

??? this may not be an error, just a misunderstanding on my part ???
Unit 02-20 2/3rds way -- Search Comparison 2
Peter Norvig claims storage for depth first search is proportional to N, but isn't it smaller. Why isn't "lowest cost first" also proportional to N. I think it's 1.5N for LCF, from 1+(b-1)/2)*N, where b is the branching factor or average number of branches per node. Norvig claims that storage for BF or UC Search is proportional to 2^N but should be more precise/general and say that it's b^N where b is the branching factor.

Unit 02-28 00:24/00:51 -- A* Search 5
Peter Norvig writes "h(s) < true cost" and writes/says in natural language "h never overestimate" when the mathematical notation should probably say <= (less than or equal). A cost estimate that is equal to the true, actual cost seems like it should be admissible, but this would only happen for a discrete cost function. If equal cost is intended to not be admissible then the natural language description should change instead of the mathematical notation to "h always underestimate."

No comments:

Post a Comment