[Termtools] some remarks

[jwulp at win.tue.nl]

> - Is TEPARLA different from last year's version?
> 
There are a few differences, this year's version uses multi-threading: different
search approaches are searched concurrently. More precisely; for a relative
termination problem SN(R/S) the following problems are considered in different
threads:

 - SN(R/S)
 - if S is empty SN(R + DP(R)) 
 - if R and S contain only unary symbols SN(rev(R) / rev(S))

Another difference is that with the new version all search procedures for
(quasi-)models for semantic labelling and polynomial interpretations are fully
deterministic. Furthermore `simpler' interpretations are tried before
complicated ones. These things makes the output of TEPARLA more predictable and
sometimes better readable.

No new techniques were added to this years version. In fact not much work has
been done on TEPARLA since September 2005. With a bug fix (for the segmentation
faults) and some more tuning to the new database the total score of the new
TEPARLA can be increased somewhat (around 10 problems). For more significant
improvements a more effective way of generating (quasi-)models for semantic
labelling or more powerful techniques for proving direct termination are needed.
Judging from the remarkable performance of Jambox this year, adding matrix
interpretations seems a good choice for the second point. For the first point I
have so far only looked into generating (quasi-)models and then sifting out
(quasi-)models based on some simple criteria. But with recursive application the
search-space grows very fast and the filter criteria were not effective enough
to actually see results within a minute of time per problem.

I will not be there next week; but I wish everyone a good trip.

Jeroen


On Mon, Aug 07, 2006 at 09:49:58AM +0200, Aart Middeldorp wrote:
> Here are some random remarks on the termination competition:
> 
> - This year's competition was exciting, but surely it can be
>    made more exciting by (first) running the participating tools
>    with a small timeout on a subset of the problems.
> 
> - Jambox is not only very powerful, it produces nice output and
>    is also pretty fast.
> 
> - Rather than adding new categories, we should consider dropping
>    existing ones which repeatedly fail to attract a sufficient
>    number of tools.
> 
> - I fully support Johannes' call for a live session at WST.
> 
> - TTT did not participate this year. The new tool TTTbox did
>    participate. From the report one may get the impression that
>    it was the other way around.
> 
> Hope to see you all at WST next week!
> 
> Aart
> _______________________________________________
> Termtools mailing list
> Termtools at lists.lri.fr
> http://lists.lri.fr/mailman/listinfo/termtools
> 
> 
_______________________________________________
Termtools mailing list
Termtools at lists.lri.fr
http://lists.lri.fr/mailman/listinfo/termtools