The Simple competitive learning rule module

The panel is divided into four sections: the Net Interface section with the buttons CONNECT , DISCONNECT and EXIT to allow the association between the network and the simulation module; the Simulation Scheduler with the buttons RUN , STEP, STOP , RND WEIGHT , EVAL , OPTION and some other tools to manage the simulations; the Training Set section with the buttons NEW and LOAD to manage the training set generation and the training set files; the forth section will show output data of the simulations and context dependent messages. The application guides the use of the panel changing the sensitivity of the panel tools according to the feasible operations at a given phase of the simulation.

The Net Interface section

CONNECT: It establishes the connection between the model panel and the current net; the net is tested to see if the structure is compatible with the simple competitive learning architecture.
DISCONNECT: It erases the connection to the net, deallocating from memory the simulation data.
EXIT: It closes the Simple simulation panel.

The Simulation Scheduler section

RUN

It starts/restarts the simulation on the current net, with current parameters and options settings. While simulation is running, the STOP and STEP buttons are sensitive.

STEP

It is used to perform a simulation step by step: when used the simulation runs for a number of steps indacated by the refresh parameter.

STOP

It allows the user to stop the running simulation; this feature is useful because the simulation time could be quite large, expecially with big nets (hundreds of neurons).

RND WEIGHT

This button allows to set random values to the arcs weight: this is useful when the user would restart a new simulation.

EVAL

After the training phase the user can evaluate the network capability to classify setting input units and pressing the EVAL button.

OPTIONS

It opens an option panel with some toggle buttons, which allows setting some simulation flags to modify both model behaviour and output features.

Report File :: selecting yes an output text file is written, while the simulation is running. The output file can be found in directory <INNEHOME>/NOTES and is named simpleReport.txt (a successive output will rewrite the last one); the output file reports different statistics on data taken after each iteration of the learning algorithm.
Eta Rule : Costant, Linear, Armonic, Exp: there are four eta decay rules: with Costant rule, the eta value does not change in the simulation; with the other rules, eta will decay according to the selected formula.
Training Set :: input pattern can be presented to the network in sequence as they are created or in random order.
Planar Visualization:: when the simulation is running in the bi-dimensional space, input pattern and weight vectors can be viewed as point or cross.
3d Visualization:: when the simulation is running in the three-dimensional space, weight vectors can be viewed as point or vectors connecting the points with the center of the distrutions set.
Dead Units:: there are four ways to avoid the problem of dead units. The user can try to combine some of them. When Init Arc Weight or Increasing Pattern are set on, the Rnd Weight button in the Simple panel is obviously disabled.

The Simulation Parameters

Initial Eta Value: It's the value used for eta by the first simulation step. While the simulation run, eta value decaies according to the Eta Rule .
Learning Cycles: It represnts the number of steps performed in the simulation.
Refresh Sequence: It's the number of steps between two video refreshes.

The Training Set section

NEW: If the net has two (or three) input neurons, it is possible to build a bidimensional (or three-dimensional) composition of gaussian distributions. When the button is clicked, a new panel opens where the user can set the distributions in the bidimensional (or three-dimensional) space.
LOAD: Loads from a training set file a sequence of example; the input neurons determines the expected format of the file; if this format is not satisfied, the load fails.

Back to index