Learning from positive examples...

Greetings...

I know c.a.n-n isn't exactly the right forum for my question, but as far as
I know, there are no NGs dedicated to machine learning in general.

Most (all?) theoretical results in learnability and computational learning
theoretic frameworks (I'm thinking especially of PAC-type stuff here)
crucially hinges on the availability of both positive and negative examples
(I'm talking about classification here, obviously)...

Does anyone know of literature pertaining to:

1) algorithms that learn from positive data only ?

2) theoretical results on the power of learning from positive examples ?


Regarding (1), I know that unsupervised learning methods can be construed
this way, but that's not really what I'm thinking of...unfortunately, I
can't specify any better what I AM thinking of...sorry. Mainly it's
theoretical results that I want to see, anyway...



Many thanks,

Fred.