Homepage of Louis Coder (Matthias Mueller)
Main page > Louis' Program to build and test (kind of) Neural Circuits







Author's website - Main page

Neuron Connection IDE

- Build and test neural circuits -

Download

./NCIDE 1.0 Beta 2.zip

.ncide Files for Beta 2 (contain circuits for various purposes, download to disk and press F1 in running NCIDE to load from disk):

- no additional files yet, but there are many sample .ncide files included in "./NCIDE 1.0 Beta 2.zip" (downloadable via the link above!).




Installation as in the following video:

Run NCIDE and press F9 on your keyboard for a short summary about the purpose, the usage, and the basic working mechanisms of the program.

More to come soon...

2022-08-15








Obsolete

./NCIDE 1.0 Beta 1.zip
[obsolete, please don't use any more,
as .ncide file format changed in Beta 2
and later]
[revised on 2022-08-16, for archiving]

.ncide Files for Beta 1 (ATTENTION: cannot be loaded by Beta 2 and later anymore, please use at least Beta 2 for saving your own circuits!):

Test.SuccessionRecognition.ncide - [obsolete] Detects successions of words just by AND-ing random points from axons of these words. Run NCIDE, load this file (or the newer version within the Beta 2 zip downloadable above) by pressing F1, and read the annotations showing up in the application window.

Test.WillToSpeak_'say hello'.ncide - [obsolete] 'The opposite' of Test.SuccessionRecognition.ncide - I think there's a uniformity in the brain, here you can see that it could work like this, long axon pathways (signal propagation times matter!) with chaotically connected dendrites of neurons which (logically) AND their input.

Test.RetinaPatternRecognition.
LearningNeuronAttempt.ncide
- [obsolete] Attempt to associate a retina image with the output of some neuron by increasing neuron input weights (as done in computer science) - doesn't work right, success depends on the right duration of the excitements (the longer, the higher the weight increase), the timing (input nodes must be excited at the same time to increase weight) and the count of input nodes - not good.

Test.RetinaPatternRecognition.
ChaoticalANDAttempt.ncide
- [obsolete] Second attempt - the underlying hypothesis of this test is that there's at least one neuron that recognizes the image FROM BIRTH because it is "by chance" connected with the "right" pixels (i. e. retina cells), only the response of its one output axon needs to be connected in a learning process - in reality, there must probably be layers of those neurons which recognize sub-parts of the retina image (pyramidal), because the count of possible retina images is too high to be detected each by one single neuron. This attempt works well. Notice, with this approach, also "fixed" images could be detected from birth (to detect dangers like predators etc.), the output of some "predator-detection"-neuron, for instance, would just need to be hard-wired in such a way that it always leads to a reaction of the individual.

Test.RetinaPatternRecognition.
ChaoticalANDAttempt.
SupportsShiftedImage.ncide
- [obsolete] Like the previous test, except that the detection of the test image is achieved independently from the location on the retina. Besides the content detection, another set of axon branches leads to a simple unit determining WHERE something is seen (here just where enough retina cells are activated, a simplification). Making the content recognition independent of the size or angle of the retina image could work accordingly.


Author's website - Main page

Privacy policy