CANT 1.8

The CANT applet should have appeared as another window. The applet version is incomplete, and doesn't allow you to write files. However, many experiments should be able to run. The code can be downloaded as a WinZip file; it should show more things and allow file writing. If you have any problems feel free to contact c.huyck@mdx.ac.uk

From the tech report on Cell Assemblies and Neural Network Theory: From Correlators to Cell Assemblies there are several experiments.

  1. Experiment 1 section 2.2.1: Correlational Learning Print Axonal Variance (A is top, B bottom, and C the middle) whenever average axonal weights are needed. set the
    1. ChangePatternEachTime on
    2. AssociationTestOn on
    3. AllowRunOn off
    4. NeuronsFatigue off
    5. CompensatoryLearning off
  2. Experiment 2 section 2.2.2: Correlational Learning
    As experiment 1, but set the learning rate to 0.05 (SetParameters LearningRate menu item. Followed by Text Entry and EnterParameter Button)
  3. Experiment 3 section 2.2.3: Correlational Learning with Normal patterns.
    Again this is similar to Experiment 1, but set
    1. ChangePatternEachTime off
  4. Experiment 4 section 3.1: Fatigue.
    Similar to Experiment 1 with Neurons Fatiguing. Set
    1. NeuronsFatigue on
  5. Experiment 5 section 3.2: Spreading Activation.
    Similar to Experiment 3 with Spreading Activation On. Set
    1. NeuronsFatigue off
    2. CompensatoryLearning off
    3. Note, that only rarely does a neuron fire via internal activation. Lower the Activation Threshold to 3 (it's 5 by default) by SetParameters ActivationThreshold, and there will be internal activation.
  6. Experiment 6 section 4.4: Overlapping CAs, using normal patterns.
    This works better than described in the tech report. The learning parameters have been set differently, and generally, two overlapping CAs are formed. Leave things at the default values.
  7. Experiment 7 section 4.4: Overlapping CAs.
    Similar to Experiment 6 with alternating patterns. Set
    1. ChangeEachTime on
    2. AssociationTest on
    3. AllowRunOn off
    Train for 8000 Cycles, then test by Set
    1. ChangeEachTime off
    2. AssociationTest off
    3. AllowRunOn on
    4. You may also turn LearningOn off
    5. Note that axonal variance is much lower than Experiment 6.

Hints for running

  1. Parameters are often changed by selecting the menu item, then entering the value in the text box, and then pressing the Enter Parameter Button.
  2. The simulation has a delay when in run mode. This delay can be modified by using the Execution SetRunDelay menu item (followed by text entry and the parameter button).
  3. You can create a new network (untrained) from the Net NewNetwork menu item. You may also read in a pre-existing net via Net ReadNetwork, which reads the network on the server specified by the Net SetNetworkFile menu item.

  5. alt="Your browser understands the <APPLET> tag but isn't running the applet, for some reason." Your browser is completely ignoring the <APPLET> tag!
    A picture of Chris

    Page last modified