1. Even though we switched to ET 226, part of the deal was that we can't use that room a couple of times this semester. This Friday is one of them. So we will meet in ET 227 this Friday. (The other time is in April.)

2. Genetic algorithms and neural nets are not direct competitors. Here's an article talking about using genetic algorithms to train neural nets. 

3. After class Weronika asked whether Robbie was a Finite State Machine. It is, but it has exactly one state. It has 243 transitions, one for each possible five element configuration. Each transition takes Robbie from its single state to that same state. It also produces as output an action to be performed. So as an FSM it has no memory, not even a finite memory. Another way to think about Robbie is as a crippled Turing Machine that operates on a bounded 2 dimensional array rather than an unbounded one-dimensional tape. It is crippled in that it can erase squares by picking up a can, but it can't write on a square. So to a limited extent it can use cans (and the walls) to help it find its way around the grid. You'll notice that successful robots run around the grid either clockwise of counter-clockwise. That means that the actions in the corners and at the walls must be consistent with each other. That's one of the things the Genetic Algorithm accomplishes. That gets the robot to move around the grid. But it doesn't have a way of moving into the interior of the grid except by following a trail of cans.

4. If you want to play with the model it's available here.

There is no accept state. You're right about the regular expression if you think of it as an expression for the possible input.