Previous: Nexus
Up: NEURAL NETWORK SIMULATION TOOLS
Next: Pittsburgh
Previous Page: Nexus
Next Page: Pittsburgh
General Information
- Name
- NSL 2.0.2 (Neural Simulation Language Version 2.0.1)
- Source
- Alfredo Weitzenfeld
(Brain Simulation Laboratory
Center for Neural Engineering
University of Southern California)
- Price
- public domain
- Target usage
- education, experimentation, research;
beginners, advanced, experts
Model
- Network Architecture
-
Designed for Leaky-Integrator Neural Networks.
Neuronclusters can be generated (single Units, Vectors, Layers)
and linked by specification of dynamical equations in mathematical
form (easy to do within C++). One set of equations serves for
a hole layer of anny type (very nice!). The form of equations is
restricted to polynomials and various kinds of sigmoid functions.
Learning may be implemented in the same way: define weightlayers
and a dynamical learning equation (Differential equation).
Missing: - No way to introduce noise into a model.
- No spiking neurons.
- Paradigms
-
- Limitations
-
Restricted to Leaky Integrator Units.
No noise.
No welldefined interface to other Applications (Input/Output).
No interface for user defined functions (These must be placed
directly at different locations of the source code.)
In principle the user may implement all the features he likes
(inluding other than differential equation dynamics)
by adding appropriate code but this needs an advanced programming
course in C++.
Implementation
- Display
-
values of units as area-level-graph, temporal graph or 2/3d-graph;
further informations about layers,units & links as text;
Possible to visualize large arrays of neurons as
quasicontinous fields (3-d-graph).
Window dump possible (Rasterfile,PostScript)
- User Interface
- shell, script (batch), GUI, mouse
- Execution
-
interactive command interpreter or scripts (batch)
batch proccessing mode somewhat restricted if large series
of expermiments required.
- Data Format
- special purpose
- Evaluation
-
- Performance
-
Something happens on the screen if you start a 200-unit application.
(Is this accurate enough ?)
System Requirements
- Hardware
- Sun
- Operating System
- UNIX
- Display
- mono (SunView only)
- Language
- C++
Installation and Documentation
- Installation
- executable, makefile (easy)
- Sources available
- yes
- Implementation language
- C++ (nearly without comments !!)
- Documentation
- no online help, incomplete manual,
examples (most without comments)
- Support
- no
- User Groups
-
Previous: Nexus
Up: NEURAL NETWORK SIMULATION TOOLS
Next: Pittsburgh
Previous Page: Nexus
Next Page: Pittsburgh