|
Technical Report 2004/153.
May 2004.
Abstract:
Examined, the succinct expression of general solutions
to inductive inference problems.
Haskell types and type classes define the properties
of various kinds of statistical model -- distributions,
function models and time-series.
This is an application of Haskell which
itself has applications, and
is almost as general as Haskell's own area of application.
Case studies in inductive inference, including
mixtures of Markov models,
state-based time-series,
missing data, and
mixed Bayesian networks,
illustrate the functional style of programming with models.
Polymorphic types, type inference, high-order functions and lazy evaluation
are all useful.
Keywords:
Bayesian networks,
inductive inference,
machine learning,
minimum message length,
MDL,
MML,
statistical models.
[TR153.pdf],
[seminar slides].
Also see
[more refs]
and later work.
|
|