Hi, Dongsong,
Here is a good reference on temporal neural networks:
[Mo94] M. C. Mozer. Neural Net Architectures for Temporal Sequence
Processing. In Time Series Prediction: Forecasting the Future and
Understanding the Past (Santa Fe Institute Studies in the Sciences of
Complexity XV), A. S. Weigend and N. A. Gershenfeld, editors. Addison-Wesley,
Reading, MA, 1994.
NIPS and UAI have always had good papers on temporal ANNs and temporal
Bayesian networks. Last year's AAAI (AAAI-98) also had a good paper by
Rosenstein and Cohen on delay space embedding by clustering.
You can find many references for time series learning using simple
recurrent networks, time delay neural networks, and Gamma memories in the
bibliography of my dissertation:
[Hs98] W. H. Hsu. Time Series Learning With Probabilistic Network
Composites. Ph.D. thesis, University of Illinois at Urbana-Champaign
(Technical Report UIUC-DCS-R2063). August, 1998.
which is online at http://www.ncsa.uiuc.edu/People/bhsu/thesis.html.
See also the ICML and AAAI proceedings for papers by Koller and
Russell on recent research on learning with dynamic Bayesian networks.
AAAI-98 is the only one I have handy; I have yet to mine the KDD literature
(pun intended).
A couple of good books to refer to are:
Kantz, H. & Schreiber, T. (1997). Nonlinear Time Series Analysis. Cambridge,
UK: Cambridge University Press.
Cover, T. M. & Thomas, J. A. (1991). Elements of Information Theory. New
York, NY: John Wiley and Sons.
=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=
Daphne Koller, Padhraic Smyth, and Leslie Kaelbling each gave some
interesting invited talks at AAAI last summer. They outlined some aspect of
learning with temporal Bayesian networks, HMMs, or other ARMA process models
(e.g., temporal ANNs). Professor Koller went over some of the DBN work in
more detail in other AAAI and ICML presentations, and Professor Smyth alluded
to some interesting dualities between temporal Bayesian networks and known
ARMA models (in the AAAI-ICML workshop on time series analysis).
=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=
As for codes, NeuroSolutions (a highly configurable, commercial ANN
simulator published by NeuroDimension, http://www.nd.com) has many types of
temporal ANNs as well as useful visualization tools and facilities for
custom development (in C/C++).
If you are interested in developing new temporal ANN simulations using
a well-tuned code base, you might try the Stuttgart Neural Network Simulator
(SNNS) at ftp://ftp.informatik.uni-stuttgart.de/pub/SNNS/ (I recently learned
of it and have not used it myself). Radford Neal's package for Bayesian
learning in ANNs at http://www.cs.toronto.edu/~radford/fbm.software.html is
very well-tuned, but as I can tell you from experience, modifying it to
simulate recurrent ANNs is a nontrivial task.
Finally, Kevin Murphy developed a code for dynamic Bayesian network
learning. This is the only "research domain" code I know of for DBNs.
Hope this helps,
Bill
=======================================================
William H. Hsu, Ph.D.
Research Scientist, Automated Learning Group
National Center for Supercomputing Applications (NCSA)
[EMAIL PROTECTED]
http://www.ncsa.uiuc.edu/People/bhsu ICQ: 28651394
=======================================================
>
> Hi,
>
> Does anybody know any belief networks literature or software that deal with
> time series problem? Or Is there any knod of neural nets which can cope with
> time series? Any suggestions are highly appreciated. Thanks
>
>
> Cheers.
>
> --
> Dongsong Zeng
>
> ********************************
> Email:[EMAIL PROTECTED]
> *********************************
>
