Source Code Download:

Echo-state network simulator (Matlab code)
New version of the ESN simulator.

Recurrent neural network simulator (Matlab code)
RNN simulator for custom recurrent multi-layer perceptron network architecture.
RNN contructors avialable for:

• Elman's simple recurrent neural ntwork
• Williams and Zipser's fully recurrent network

RNN training algorithms avialable:

• Backpropagation
• Backpropagation through time
• Real-time recurrent learning
• Extended Kalman Filter - backpropagation through time approach for linearization
• Extended Kalman Filter - Real-time recurrent approach for linearization

Simple language generator (Java)
Can be used for generating simple regular and context free languages, and also for Monte-Carlo estimation of language entropy. See short description how to use.

Fractal prediction machine tool (Matlab code)
Next-symbol prediction tool based on fractal representation of histories of input symbols in RNN-like (Recurrent Neural Network) state space. See fpmdemoreber.m orfpmdemolaser.m for more detailes.

Echo-state network simulator (OOP Matlab code)
Obsolete version of the ESN simulator. Was used for playing with experiments described in Herbert Jaeger's technical report located at Herbert Jaeger homepage. See demonstration files esndemo*.m for more detailes.

Data Sets Download:

Grammar descriptions for Simple language generator:

• Reber and Extended Reber (simple regular languages 6 symbol alphabet)
• Language anbn (the simplest context-free language, 2 symbol alphabet)
• Triangle automaton (simple regular language, 4 symbols alphabet)
• Recursive languages (based on Christiansen and Charter's article, 17 symbol alphabet)
• Deep recursion  language (simple context-free language, 4 symbols alphabet, max recursion depth 10)
  

Symbolic time series:

• Laser (10000 symbols, 4 symbol alphabet)
• Feigenbaum (5.0e5 symbols of Feigenbaum sequence, 2 symbol alphabet)
• Reber (1.0e6 symbols)
• Triangle automaton (1.0e6 symbols + 1.0e6 symbols representing states)
• Recursive languages1 (based on Christiansen and Charter's article, too short sets)
• Recursive languages2 (3 languages, 17 symbols, 5.0e5 and 5.0e4 symbols in training and testing sequences)
• Deep recursion  language (4 sets, various recursion depths and word frequencies)
  

Real-value time series:

• Lorenz chaotic time series
• Mackey-Glass chaotic time series
• Laser time series