Addressing three important problems.
- defining learning problems
- showing specific algorithms work
- show these problems are fundamentally hard
The tools that used to analyse learning question also are the tools to
analyse the Algorithm in Computing.
Three important thing to a learning algorithm:
$\rightarrow$ Time
$\rightarrow$ Space
$\rightarrow$ Sample (data sample) or generaliazation
Inductive learning
Inductive learning : Learning from sample
- Probability of successful training (it might not work)
- Number of examples to train on
- Complexity of hypothesis class
- Accuracy to which target concept is approximated
- Manner in which training examples presented (batch)
- Manner in which training examples selected
Select training Examples
It matters that how we select training examples when a learner is
needing to learn.
Learner/Teachers
- Learners asks questions of teachers. given sample x, ask c(x)
- Teacher give examples to help learners, teacher chooses x, tells c(x)
- Fixed distribution, x chosen from $D$ by nature
- Evils asked questions
to be continue..