Fading:  When a Bayes net is used in an environment that is changing, it may be desirable to have it adapt to its environment by constantly learning, in a process that weights cases it has seen recently more strongly than those seen in the far past.  The process of discounting knowledge learned from the past is called fading.  more info

Finding:  A finding (also known as “evidence”) is a value for one of the nodes (i.e. variables) of a Bayes net when it is applied to a particular situation.  more info

Findings menu:  A discrete node’s findings menu is obtained by right-clicking on the node.  It lists the states of the node, and may be used to enter a finding.  more info

Findings node:  A findings node is a node which has a finding, or which we know will receive a finding before doing belief updating.  Compare with target node.

Function table:  When the relationship between a node and its parents is deterministic, rather than probabilistic, then instead of a CPT a node may have function table, in which each row corresponds to a configuration of parent values, and the row provides a single output value for the child node (i.e. the table is a function mapping parent node tuples to child node values).  If a function table is converted to a CPT, then each row of the resulting CPT will consist only of zeroes, with a single 1 (or 100%) positioned at the state that was the function table's value for that row.  more info

Home folder:  The Netica home folder is the folder that the Netica executable "Netica.exe" appears in, as described in Installation.  Normally each version of Netica on your system has its own home folder.  When Netica is running, you can discover which folder is the home folder by choosing Help → About and then looking at the last line of the Messages window. Also sometimes called the Netica home directory.

Hugin:  A quality program available from Hugin Expert A/S which works with Bayes nets much the same way Netica does.  Netica can read .net files produced by Hugin versions 4, 5 or 6.

IDname:  An IDname is any word that starts with a simple letter (a-z or A-Z), is composed only of simple letters, digits and underscores (_), and is 30 or fewer characters long.  It must not contain any spaces or punctuation.

iff:  iff means “if and only if”.  So the expression “A iff B” can be interpreted to mean “if A then B, and if B then A”.

Ineffectual link:  An ineffectual link is one in which the CPT table of its child node contain the same probabilities independent of the state of that link's parent node.  In other words, the table contains replicated chunks, with an identical chunk for each state of the parent node.

Ineffectual links may be removed without effecting inference results at all.  When a link is first added, it is added as an ineffectual link, and remains that way until the CPT tables of its child node are modified.

Netica can find and display all the ineffectual links in a net.

Informationally independent:  A variable X is said to be (informationally) independent of another variable Y if obtaining knowledge about the value of X does not change your beliefs about the value of Y.  If X is informationally independent of Y, then Y is informationally independent of X.  See also conditionally independent.

Informational link:  Any link entering a decision node is known as an informational link, and indicates that the decision maker will know the value of the parent node when he must make that decision.

Invertible node:  An invertible node is a deterministic node whose relationship with its parents is invertible.  That is, the value of any one of its parents may be expressed as a function of it and the rest of its parents.

Joint distribution:  The joint distribution provides a probability for every possible outcome (i.e. every element of the cartesian product of all the given variables).  Since one, and only one, of these outcomes will eventually occur, the joint distribution probabilities add to 1.

Joint probability:  A joint probability is a probability from the joint distribution.

Junction tree:  A junction tree (also known as a “join tree”) is the internal structure that Netica uses for belief updating.  Netica compiles a Bayes net or decision net into a junction tree for efficiency.  more info

Latent node:  A latent node (also known as a “hidden node”) is a node in a learned Bayes net which does not correspond to any variable in the input data.  That is, the variable was created to more easily express the relationships between observed variables.

Leaf node:  A leaf node is a node with no children.  See also barren node and root node.

Likelihood finding:  A likelihood finding (also known as “virtual evidence”) is a finding with some uncertainty attached.  Compare with negative finding and positive finding.  more info

Link:  A link (also known as an “arc” or an “edge”) is a connection between two nodes indicating dependence, and is usually drawn as a line with an arrow at one end.  more info

Link reversal:  Link reversal is the process of changing the direction of a link, and then making necessary changes to the rest of the net so that any subsequent inference results will not be affected.  However, it may result in less efficient inference (or occasionally more efficient), since extra links may have to be added (removed) to maintain the global relationships between the nodes.   more info

Link structure:  The link structure (also known as the “topological structure” or “dependency graph”) of a Bayes net or decision net is just the graph structure of the net.  In other words, it consists only of the node names and links, but not of any other information about the nodes or the particular relationships the nodes have with their parents.

Loop cutset:  A loop cutset for a net is a set of nodes such that if they are all removed from the net (along with all links involving them) there will be no loops in the net.  See also cutset.

Markov blanket:  A set of nodes B is a Markov blanket of node X, if given findings for all the nodes of B, X is independent of all other nodes in the net.  If it is the minimal such set, then it is called the Markov boundary (see below).

Markov boundary:  The Markov boundary of a node X is the minimal set of nodes for which any set of positive findings can be supplied and X will be conditionally independent of all other nodes in the net.  For a Bayes net, this is X's parents, X's children and X's childrens' other parents.

A set of nodes Y can also have a Markov boundary.  If positive findings are obtained for the Markov boundary nodes, then findings for the Y nodes will not provide any additional information about any other node in the net and vice-versa.

Netica can find and display the Markov boundary of a node or set of nodes.

Markov network:  A Markov network is an undirected network in which the nodes represent variables of interest and the connections between them represent probabilistic dependence.  Netica converts a Bayes net into a Markov network as an intermediate step in producing a junction tree.

Mean value:  In calculus, the mean value theorem states, roughly, that given a section of a smooth curve, there is a point on that section at which the derivative (slope) of the curve is equal (parallel) to the "average" derivative of the section.  More Info

Messages window:  The Messages window is a window that Netica uses to communicate textual information to you.  You can open it any time using Window → Messages, and you can copy and paste information between the Messages window and any text file.  Sometimes if you make a minor mistake, Netica just beeps and places a message there.  more info

Missing data:   If some cases have values for a certain variable, and others do not, that is known as missing data.  more info

Moral graph:  A net in which each node has all its parents connected to every other of its parents (i.e. there are no “illegitimate” children).  Moralizing a net consists of adding the required connections between parents of a node for all the nodes (i.e. “marrying” them).

Most Probable Explanation (MPE):  The most probable explanation, or MPE (also known as “maximum a-posteriori probability” or “MAP”) is a set of values (one for each node) that is the most probable configuration, given the net’s current findings.  Since there may be several configurations with the same probability, the MPE may not be unique.  more info

Multiply-connected:  A multiply-connected network is a directed or undirected network which has more than one undirected path between some pairs of nodes.  In other words, it has loops.  more info

Multipurpose selector:  When you click on the multipurpose selector, , of a node dialog box, you will be presented with a menu of node properties that can be edited in the box below the selector.  more info