Animation of AI engine states

When the output from the web cam reaches the OVideo AI engine it is processed within the nodes of the matrix.

As a consequence each node acquires a certain state which changes as the interdependencies between the connected nodes adjust their mutual relationships.

In the animation below the AI engine parameters are a follows:

Main matrix rows and columns, element matrix rows and columns: 25
block factor, number of in- and output regions, active in- and output region: 1
connection depth: 2
connectivity in %: 10
high end of element number range: 50000

15 cycles were captured, which in this case meant an operation lasting just over 5 hours (how long a cycle takes depends on the CPU). Altogether there were 325 images, capture frequency was one image per 60 seconds.

The size and colour of each node reflects its affinity with the node it is connected to, where the larger the size and the more towards the blue and/or white from the black and/or red the colour the higher the degree of affinity. In other words, the smaller and darker a node looks like the more distinct it is, and the larger and brighter it is the more it is part of a cluster of nodes. What and where in the matrix these other nodes are can be gleaned from their similarity in appearance (the connections do not follow a systematic layout, so the members of a cluster can be anywhere in the matrix).

Note: As far as the appearance of the nodes in this animation is concerned, it does not matter what the input was, the graphical depiction is only an interpretation of the affinity parameters. The output of the matrix on the other hand is a function of the node states and they are based on integers. Therefore this animation could have been made in any shape or form, to suit any occasion; the affinity values could have been translated into sound frequencies for that matter, even though there was no sound input in this case. Instead of little ovals for each node more sophisticated visual effects could be used as interpretation, the only limit the real-time processing capacity of the system (it therefore helps if the entire matrix was in distributive mode).

© Martin Wurzinger - see Terms of Use