|
We have seen the Minimum Message Length (MML) and other treatments
of "simple" hypotheses over
"simple" data such as finite choices,
e.g. multi-state,
and continuous values,
e.g. normal distribution, etc.
It turns out that structured hypotheses
can be formed from combinations of simple hypotheses
plus a component to describe the particular structure.
A few examples:
hypotheses
|
simple hypotheses
|
structured hypotheses
|
MMLFP:
ACSC03,
JFP,
ACSC06,
II
1.0
1.1
|
discrete
|
continuous
|
Sequences, time series, grids, images,...
|
unsupervised
|
supervised
|
|
finite, multistate
|
enumerable, integers
|
normal
|
... |
Markov models,
HMM+PFSA,
BioInformatics,
Segmentation
|
classification, clustering,
mixture model,...
in series,
factors,...
W+F'87...
W+B'68
|
... |
decision tree,
decision graph,
...
|
... |
Bayes net,
hybrids
|
|
|