Closed Twenkid closed 6 years ago
It's a POC for 1D (line) operations. It does have first-order feedback: prior pixels, and there will never be a full set of feedback orders: number of levels can grow indefinitely. POC means a basic demonstration, not a complete or competitive system.
"Never full set of fb" - this is an easy excuse for not having a feedback dynamics, which however is the essential in ANN and would be in CogAlg, too. As defined, the first-order feedback is not really convincing as a feedback either.
The process of adjustment of the filters is the "intelligence" (or "learning" in ML), not that there is some kind of traversal of the input data, some processing which matches the elements to each-other ("comparison") and "some" feedback", which however is not elaborated. In ANN it is, such as the back-propagation, SGD, ...
IMO it'd be a POC if it produced and emerged into something that's not obvious.
PS. What passes as a "POC" to me is your coding style, the tuple-formation; vP, dP classes, recursion, the skeleton of naming conventions and functions, the traversals. However to me it's not a prove of the algorithm as incremental, discovering incremental patterns etc. In order to do, it needs to have at least two layers and a feedforward-feedback dynamics which is showing some kind of objective results.
Do it, Todor. As it is, your nit-picking is an easy excuse for wasting your life.
Boris, you now call the 1le - line_POC, but I don't think it's clear how it proves your concept, without more elaboration and without being possible to test it for recognition as you say; also without the required feedback and with fixed filters.