Interactive super-logic lessons for Neural network.
in REAL TIME

Apr96, June97, Feb98, Mar98

NO use of unknowns of human brain in definitions of basic ideas & values:
net LESSONS become rigorous logic & philosophic definitions;
net EXACTNESS humans may crudely,unhappily IMITATE.

"Remember only the last trial synapse connection from a hunger stimulation to a response after the hunger stops".

General background(above)(8 pages, 3-6k gifs) This script implements the 'imitation' described there: a 'copy synapse' between a 1st and 2nd nerve is automatically produced(learned) by initially stimulating a 1st nerve(seeing a book) for the 'sensitization' time(3 agings: www), then stimulating a 2nd nerve(hearing 'book'), then stop the 1st stimulation(hide book), then stopping the 2nd(stop saying 'book') - as if the 1st(seeing book) caused the 2nd(hearing 'book'). The net is programmed to add a 'copy synapse' between the 1st and 2nd so seeing the book causes the same internal stimulation as hearing its name. Learning to say or write 'book' when it is seen, is another lesson. A 'inhibitionn synapse' is made by stoppin stimulation of the second nerve while the first is still stimulated - as if the first stopped the second as will now be done(imitation). A random process may be required to practically solve a problem. Clearly, even a random process may be the result of 'imitation' - imitate a coin toss to choose.

Input lesson Description :

• Seek & join: '001sw002sww' creates a trial copy type synapse '01C' from chain head nerve 1 to chain tail 2 see '2C1>2'. (nerve 001 is stimulated for 3 aging periods and 002 for only 2 agings). ('001sw' makes 'nerve display' show '1s1i0a1'. 's1' is 1 stimulating connection. 'i0' is 0 inhibiting connections. a1 is aged by 'w'.). This is like imitation: since nerve 2 is stimulated after nerve 1, -ANN gets a synapse that does the same thing.
• • This 'trial' synapse connection between any 2 nerves is implemented in javascript by a simple loop over all nerves; it could be efficiently done biologically by the 'seek & link' process: instead of having all nerves have branches with potential synapse connections to all other nerves(most branches are redundant), the brain is assumed full of one set of 'seek & link' nerves that can supply any connections as needed. Each 'aging' here may correspond to a 'brain wave' where the more longly stimulated 'www' nerve trial stimulates neighboring 'seek& join' nerves which intern stimulate their neighbors throughout the brain until a more recently stimulated and possibly newly relaxed (sensitized 'sww' or 'swrw') nerve is joined. To keep only one chain of 'seek & join' nerves, I propose a 'back propagation' from the second to the first. This pruned compound trial connection is kept for only a short time(short term memory?) unless it is quickly reused after a successful connection(original stimulation off, on, off). A count is kept of all current stimulating connections and all inhibiting synapse connections. The nerve is stimulated only if there are no active inhibitions and at least 1 active stimulation(a copy synapse from a stimulated nerve).
• Sensitization: '02sw' chooses nerve 2 for sensitization start with age 1. Full sensitization requires another aging. To allow a copy type synapse, the 2nd aging may be immediately preceded by a relax 'r'. This follows inverse nature of negative feedback net: trial stimulate unstimulated nerves and trial inhibit stimulated nerves; have active or sensitized nerves the target.
• Trial synapse: use next available synapse.
• 's' and 'r' apply to only preceding listed nerve.
• 'w' aging ages 1 unit all nerves before it and causes a scan for age 5 nerves to synapse to age 2 nerves. It also resets the age 6 or higher.
• '/' pair bracket instructions for step-by-step, slowed monitoring.
• Reward is relaxing head of seek nerve chain.'01rw' rewards '01' for its long stimulation of age5 by destimulating it. During the last 'w', this causes a permanent synapse connection from '01' to '02'(synapse: 01C01>02 changed to 01c01>02 otherwise failure makes synapse state 'C' into unused 'u')
• 'Input lessons' given are examples that you may retype. Click the Input button for net learning process.
• code: One can exercise the net by changing the input lesson in 'form' box: 3 digit nerve name, s timulate, r elax and w ait(' 'stops input. ). Save new input lessons by 'select/copy/paste'
• 
  I will integrate new ideas with credit into my files.

Net LESSONS become SUPER RIGOROUS logic or philosophical definitions and proofs!

DISCUSSION

Learns any t/f value logic formula = nor-formula:
For nor-formula extendability, a nor-formula that preforms a desired digital transform must be pre-constructed with extra nerves and synapses beyond those required for the transform. These extras may be called 'extendability structure' and the others 'desired structure'. Thus each desired synapse between nerves x and y must be instead a nerve(e) with 'copy' type synapse from x and the desired type synapse to y. Also nerve e must have an 'inhibit' type synapse from a unique nerve, so e may be separately inhibited as necessary to later make new synapses from or to x and y or other desired nerves. For example, if x had only a simple copy type synapse to y and one wanted to teach the net to have nerve z be a copy of x, when one stimulated x, y would also be stimulated and 'seek and link' nerves would be generated from both, not just from x. But with e between, it could be first inhibited so stimulating x would only cause 'seek and join' from x since y would not automatically be turned on..

Simulates human learning: "Remember only the last trial synapse connection from a hunger stimulation to a response after the hunger stops". Learning to learn as a child can be automatically using only those taught rules that are most often successful. The more general simple rules will more likely be kept than specialized ones because they need many learned exceptions.

Input lessons are super-rigorous logic and math: Computer science shows how to make counters from 'flipflop's. Counting and 'computer arithmetic'(not the old Peano arithmetic axioms requiring an n+1 for any n) can be thus defined for -nn. This -nn 'computer arithmetic' and -nn logic are not limited by Godel's incompleteness arguments: see INTERACTIVE CONUNDRUMLESS Computer cycling:(1page display/3page netscape2 JAVASCRIPT, 15min. study) All computer programs have a finite cycling period; programs cannot compute their own period or need to: questions Turing's, Godel's and Penrose's theories for real computers.
INTERACTIVE CONUNDRUMLESS Russell's Paradox:(1page display/2page netscape2 JAVASCRIPT, 5min. study) Computer just cycles name of catalogue of books that do not cite themselves.

Consistent with basic physical evolution of increasing entropy faster: -ANN reward is stopping the stimulation after trial mapping of that input to the output. This causes a change in the physical environment that probably would have remained the same; since any change increases entropy, the entropy was probably changed sooner as a result of the -ANN; and since nothing else can use up the same free energy already used for this entropy increase, this -ANN has this automatic 'success reward'. It should be noted that long term objectives make future responses more likely successful. And this is only predictable if the environment is suffiently repeatitive But for a completely chatoic universe, these repetitions evolve into otherwise unreachable states. And the -ANN learnes those NF's that change that repitition, so it increases chaotic entropy sooner and uses up repetition or order to do it. Thus -ANN or similar human action is a necessary part of the universe's chaotic self causation.

Input lessons may be super-rigorous values definition: for general background(above)(8 pages, 3-6k gifs) Following the general physical evolutionary goal of increasing entropy before others do requires taking advantage of the efficiency potential in a society of similar individuals or other -nn, that may be due to specialization with honest trade and the averaging unpredictable results with honest insurance. Consciousness is simply the use of one's learned model of one's effect in the universe - the more accurate the model, the greater one's entropy effect.

The ASCII output is switched on by stimulated and uninhibited nerve 021. Nerve 022 outputs new lesson header. Nerves 31 thru 38 are the binary ASCII code unescaped at end of lessons TEXTAREA. This TEXTAREA may be pasted into from a copy of e-mail from other sourses so different neural nets may communicate. Special e-mail options could make this automatic on opening the neural net page. I'm investigating use of public storage areas like forums or 'HyperNnews' for neural net communication, also modification of common search pages and use VBScript 'CreateTextRange' and ''textrange.text' for -ANN to 'surf the web' and automated bank account, trading accounts, bank accounts to use the information and become an active 'artifical citizen'.

SUPER LOGIC:(2page display, 8k-js, 1 to 8 hour study) Old - no textoutput, but slightly different.

Integer based calculus. 30 min. study (10k) General simple example of explicit finite nor-formula interpretation of differentiation(relative rates of change)