This sounds like exactly kind of problem that the technologies used in
Watson are designed to solve.
No clear path to an answer; A number of different ways to evaluate the
incoming signal; The final answer or answers can be ranked by
probability of truthfulness; You would like the system to improve itself
as it experiences different situations. (01)
On 24/02/2011 10:11 AM, Bruce Musicus wrote:
> One barrier that we are dealing with currently is that applications that one
> might think would be "obvious" for ontologies appear to be quite difficult
> to implement in languages like OWL due to difficulties with getting the
> details right. An example is a signal detection and classification
> application we've been playing with for a couple of months, trying to see if
> we could use ontologies to describe classes of radio signals that might be
> of interest, classes of noise that might be encountered, and then reasoning
> from the ontology to suggest signal processing strategies that are
> recommended for a particular classification task. Although OWL 2.0 has
> added ranges for scalar variables and pattern match to strings, it is
> difficult to do any computation within OWL as part of the definition of the
> signal class. This means that we have to augment OWL with other languages
> and rule systems (e.g. JENA, MATLAB, etc.). Furthermore, signals in real
> life are not always so clear cut, so we need to reason with uncertainty, and
> thus we need to add Bayesian-like conditional probabilities to the ontology
> and then be able to reason with these probabilities. This has led us to
> look for some sort of hybrid Bayesian Network/ontology strategy, but we're
> having trouble finding released and supported software that might succeed at
> this task. As we continue to pile on more languages and tools to handle
> additional aspects of the problem, the resulting Rube Goldberg system
> becomes less appealing and more difficult to sell as a practical and useful
> idea. (02)
> Of course, we're just starting to look at this area, so we are still
> relatively ill-informed and open to suggestions.
>
> I am curious about the idea that a language like OWL might be suitable for
> describing a project's requirements. UML has had reasonable success at
> providing a conscise description of an object-oriented program, though it is
> still a bit controversial due to the cost of the tools and the extra labor
> required to create the UML model and then maintain it as the code evolves.
> Are there projects that could be better managed using OWL instead of UML?
>
> The best applications of ontologies still seem to me to be knowledge
> representation, especially when linked with natural language understanding
> and free-form or semi-structured text. Applications to bioinformatics seem
> well developed, though I am still trying to understand how these
> applications deal with the inevitable contradictions that you find in the
> literature.
>
> -- bruce musicus, Aptec Group, NH, USA
>
> -----Original Message----- (03)
_________________________________________________________________
Msg Archives: http://ontolog.cim3.net/forum/ontology-summit/
Subscribe/Config: http://ontolog.cim3.net/mailman/listinfo/ontology-summit/
Unsubscribe: mailto:ontology-summit-leave@xxxxxxxxxxxxxxxx
Community Files: http://ontolog.cim3.net/file/work/OntologySummit2011/
Community Wiki: http://ontolog.cim3.net/cgi-bin/wiki.pl?OntologySummit2011
Community Portal: http://ontolog.cim3.net/wiki/ (04)
|