Glossary of Definitions used in Paradigm Methods and Parameters

by Martijn Meeter & Jaap Murre

There are several object types in Walnut. All have their own characteristics, and all can have an arbitrary number of parameters specified by the author of the paradigm used. For an exhaustive list of rules regarding name-giving in Walnut/Nutshell paradigms, see Name Conventions for Paradigm Methods and Parameters.

Below, in this document:

Top of this document.

Nodes and Node Parameters

[Back to contents]

Node definition

Model neurons are called "Nodes" in Walnut (this is common terminology in connectionism).

Standard parameters

Nodes have typically connectionist parameters such as an activation state, and an output state. These states are captured in parameters with names such as "Act" and "Fires". There are a few conventions governing the namegiving of these parameters in Walnut; these are discussed in the "Parameter Conventions" document, see Name Conventions for Paradigm Methods and Parameters.


"Clamp" refers to freezing the state of the node. The state of the node will, however, continue to influence weights and other nodes.


"Deactivate" refers to freezing the node in that state. A "deactivated" node, unlike a clamped one, cannot influence weights or other nodes.

Layers and Layer Parameters

[Back to contents]

Layer definition

"Nodes" are always organized in a "Layer", a group of nodes that belong together. Layers are visualized in Nutshell as two-dimensional sheets of nodes with adjustable width and height. In Walnut, however, layers are implemented as a one-dimensional array of nodes.

Standard parameters

Layers may implement a whole range of parameters. Standard parameters are the "Width" and the "Height" parameter that embody the size of the layer: they both code for the number of nodes in the layer (Width * Height) and for the physical appearance of the layer in Nutshell.

Other parameters

Other parameters are at the discretion of the paradigm builder. There is a convention governing hard-k-Winner-Take-All dynamics, however.

Connections and Connection Parameters

[Back to contents]

Connection definition

Every two nodes can be connected or not. If they are, there is a connection between the two. In many paradigms, a node cannot be connected with itself.

Connection parameters

Connections typically implement a Weight that codes for the "strength" of the connection. Another parameter that exists in most paradigms is "Conn Exists". This is the parameter that shows whether the two nodes are in fact connected or not.

Tracts and Tract Parameters

[Back to contents]

Tract definition

Connections are organized in "Tracts". These group together the connections sent from nodes in one layer to nodes in another layer (it is also possible that the sending and receiving layers are the same; in that case, one speaks of an autoconnection). A tract only groups connections that go one way - if between two layers there are connections going back and forth, then there will be two tracts between those layers - one for the forward connections, one for the backward connections.

Tract parameters

Like layers, tracts may implement a whole range of parameters. These are at the discretion of the paradigm builder, except for the weight maximum and normalization parameters.


[Back to contents]

In addition to an arbitrary number of parameters, a module contains either one or more layers or one or more modules. It may be empty, but it may not contain both layers and modules.


[Back to contents]

A model is a module, but a model may not be contained in a module. It can, thus, serve as the top-level module. It is not a workspace.


[Back to contents]

For any object, subtypes may be defined. For example, we can have in one paradigm, a CALM module, an output module and an input module. Each of these may have different sets of parameters and internal architectures. Connecting two containers may result in different tracts being inserted, depending on the subtypes. Subtypes are currently not implemented.


[Back to contents]

A workspace can contain zero or more models, modules, layers or other objects. It is equivalent to a window in the Nutshell browser. A new workspace is always created when a paradigm is opened.

Standard Methods

[Back to contents]

These are methods that can be specified by the user. An example would learning rules for a connection or tract. These are currently not implemented.

University of AmsterdamUniversity of Amsterdam
Department of Psychology
Page last modified: 2008-09-29. Validate html.
Copyright © 2000-2009 NeuroMod Group. Send us Feedback!