@Klangraum

2 years ago

Basically, I would call the abstraction of such a network a filter design. After a while trough the network you will find a path of lowest resistance between given by input and output. Maybe like a potential difference, say like in a circuit where the path is the least resistance in a conductor, or a flash in an ionized field in a thunderstorm, or the simplification of a complicated mathematical expression (which would be an ideal path compared to the physical phenomena which alway produce some jitter). I would also compare it to a slightly more complex form or different representation of a neural network, where multiple neurons form a functional group (like a cellular automaton). But I don't see any benefit in computational problems that still need to be serialized, even if we were talking about a complete filter frame (out of a series of frames = motion). Even if we hope that a single frame is available immediately, it will not be computationally feasible. Perhaps quantum computers would come into play here, but that again raises new problems such as statistical accuracy. If we knew all possible initializations in the filter structure (which would be needed to filter anything), we would need a consistent memory and feed it in somehow, which also takes time. (This would be comparable to storing the weights in a neural network) How are we supposed to describe the individual components nodes, connections, memory, clock (which would be frame and sub-frame cycles, which would actually work as binary counters, but...) only with binary patterns, if we use these cellular automata as a basis? This description, too, has to be stored in a memory that would then describe itself with itself. Isn't that a paradox?

4 |