Cram2 Flashcards
Why is Planning Different from Search?
- Search generates all successor states.
- Search wants to find a sequence of actions leading to a goal. Planning looks for a description of a plan.
A Plan consists of
- a set of partially ordererd < plan steps
- a set of variable assignments x = t
- a set of causal relations
Complete Plan
Every precondition of every plan step is satisfied
Consistent Plan
If Si < Sj, then Sj -< Si and if x = A, then x != B
P(A or B)
P(A) + P(B) - P(A and B)
P(A | B)
P(B)
P(A and B)
P(A | B) * P(B)
Bayes: P(B | A)
P(A)
P(Y | X,E)
P(X | E)
d-separation
- can be computed in polynomial time
- is incomplete, that is, not every conditional independence is detected
- is nevertheless sufficient for a number of inference algorithms
Goal of Learning
Optimize future behavior on the basis of the history of percepts, actions, and knowledge about the world.
Supervised Learning
Both the input and the correct output are available to the learner
Reinforcement Learning
While the correct answer is not available, there is feedback in terms of rewards and punishment
Unsupervised Learning
There is no indication of what the correct output is
Every propositional formula (Boolean function) is representable by
a decision tree
Evaluating a Learning Algorithm
Keeping the training and test sets separate is crucial!
Strategy of the Current Best Hypothesis
Only consider one hypothesis at a time. If there is a new example which is inconsistent with the hypothesis, then change it in the following way:
- generalization: make the extension bigger for a false negative example
- specialization: make the extension smaller for a false positive
Some problems with the current-best hypothesis:
- All previous examples need to be tested again.
- uncontrolled backtracking
Decision Lists
Decision lists (DL’s) consist of a number of tests, which themselves consist of a conjunction of a bounded number of literals. If a test is successful (all the literals are satisfied), then the DL tells us which value to return. Otherwise, the next test is tried. A bit worse than DT.
Advantages of Neural Networks
- Parallelism -> speed
- robustness
- graceful degradation
- inductive learning
Feed-forward topology
DAG
Recurrent topology
arbitrary directed graphs
With 1 hidden layer
every continuous function is representable
With 2 hidden layers
every function is representable
Net is too big:
Overfitting
XOR is
not representable when using a step function
What are NNs good for?
attribute-based representations, in particular also for attributes with continuous values
Minuses of NNs
no transparency
Bad at using additional knowledge
P (occ_x,y)
Initialize the occupancy values of the grid to 0.5, that is, initially the robot has maximum uncertainty about the occupancy values.
Markov Assumption
refers to conditional independence, e.g. a sensor reading only depends on the current location and the occupancy of the cells
Weak AI
machines which act as if they were intelligent
Strong AI
machines with a real (human like) conscious mind
