Dear all,
We are pleased to announce the release of PRISM 1.11.
PRISM is a logic-based probabilistic language, which is expected to be
easily learned and used by people familiar with Prolog. Its programming
system provides utilities for statistical learning and other probabilistic
tasks. These utilities work for discrete BNs, DBNs, HMMs, PCFGs,
and other less-known models, and they are even applicable to totally
unchallenged models. This feature frees us from labor of complicated
derivation and implementation of model-specific algorithms.
The new version is shipped with new key features and further performance
improvements as follows:
[New Key Features]
- variational Bayesian learning
- top-N Viterbi computation
- deterministic annealing EM algorithm
- data-parallel EM learning
[Performance Improvements]
- speed-up on learning by full inter-goal sharing (1x-15x faster on EM
iterations; 1x-5x faster on overall learning)
- memory use on tabled search much reduced (thanks to B-Prolog 7.0)
- significant speed-up on non-tabled search (thanks to B-Prolog 7.0)
For more details, please visit the PRISM website:
http://sato-www.cs.titech.ac.jp/prism/
Best regards,
Yusuke Izumi
--
Izumi, Yusuke / [EMAIL PROTECTED]
_______________________________________________
uai mailing list
uai@ENGR.ORST.EDU
https://secure.engr.oregonstate.edu/mailman/listinfo/uai
