Modèles Informatiques du Langage et de la Cognition - MILC
Artificial Intelligence Group
Computer Science Department
Français
Research
Papers
Home
Delta Neural Network Simulator
This neural network simulator has been developped with the support of
the SNCF. It is dedicated to time series prediction and provide a
great number of functionalities:
- Various statistical models (AR, TAR, ARMA, ARIMA and ARARMA).
- Various connectionist models (MLP, TDNN, ME, HME, Elman, Jordan, LFN, GLFN, NARMA, delta-NARMA).
- Various units (sigmoid, linear, epsilon-NARMA, epsilon-ARMA, LFU).
- Numerous data analysis tools (units activation/output/error plot, error distribution, weight evolution,...).
- Exogeneous variable handling.
This simulator is based on the concept of virtual error
(Bonnet et al., 1996). Its main specificity is that
it can easily deal with modular networks - i.e. networks used as units
of a higher level network. It thus allow the implementation of
delta-NARMA neural networks (Bonnet et al., 1997).
This is a snapshot of dnns v2.2. Click on the image for a larger snapshot (92K)
dnns v2.2 has been developped on Sun workstation (Solaris) with X11 release 6 and Motif 1.2.3. Nevertheless, it also works on HP-UX, Sun OS 4.1.* and with X11R5.
Last changes March 7 1997
Page created and maintained by Denis Bonnet (bonnet@inf.enst.fr)
Some statistics about this server are avalaible on my pages