incidentally, i just read a paper yesterday by a group at CERN looking at violations of time-reversal symmetry (the T in CPT) in neutral-kaon experiments. because they had to look at large numbers of particle track data (over 1 million) they used a neural network trained to search for particular sub-atomic reactions in the data.
But i dont think thats what you mean by relativistic neural nets....
Doyle I helped type up a paper on Monte Carlo methods of measuring quark events in a linear accelerator. The point of neural network in that situation was to recognize the pattern of I believe down quark patterns in the millions of collisions. It has nothing much to do with the relativity of gravity in a local large mass.
Doyle To understand in some sort of way what is the relativistic nature of neural networks I'll give the example in vision of seeing color. We see that instantly. That is a characteristic of parallel processing. About 100 million neurons in the retina respond at once to stimulation by light quanta. Edwin Land demonstrated in a famous Mondrian patch experiment that seeing color in a landscape is dependent upon the influence in a three dimensional vector space of all the colors present. The three dimensions are the three types of cone cells in the retina. And one cannot see a color unless it is in a field of colors. So a color is relativistic. There are many examples in vision of the dependence of the response of the neural network to the overall response of the network. Individual cells in a network have little impact as opposed to a sequential computation which fails if a single element fails in the logical chain of algorithmic operations.
Doyle The brain consists of thousands of modules of neural networks interconnected by massive amounts of cell processes called dendrites and axons. Without this interconnection of the networks the mind does not work. This process of interconnection is the relativistic nature of the mind. It is not about complete relativity of concepts. In other words if you say the sky is blue, I must base my concept of the sky upon yours, but there are fundamental aspects of language which are best understood in these terms. For instance in linguistics, prototypes such as metaphors are best understood as activations of a neural network. In the technical jargon, State Space (the network) vectors or activity of many neuron "computations" projecting through the brain. This sort of thing gives us direct insight into properties of language, why children learn language better than adults, how language requires kinds of modules, how to think of learning, etc. It constitutes a revolution in what we mean by consciousness. It overthrows the long standing dependence in our culture upon logically constructed thought.
Doyle Another area which is ripe is to reintroduce feelings into human culture. There are medical arguments concerning neural networks that demonstrate that no one can withdraw feelings from "rationality" without seriously degrading the capacity to make judgements. This overthrows moral conventions in English, and American culture about feelings.
Doyle Finally such things in my opinion allow us to tear apart and reconstruct our attitudes toward disabled people. We can jettison the bigotry that defines "norms" of behavior. One can see the precursor of that sort of thinking in Stephen Jay Gould's book's on the IQ. But Gould lacks the insight that knowing how the brain works provides.
Doyle And also I point out that neural networks work in a completely different manner than the rule based "universal grammars" of Noam Chomsky. Since Chomsky's rules have so far not been found in the brain, the case for rules of the brain grow shakier. It seems possible given the instinctive actions of animals, but there may be appropriate ways understanding these behaviors given neural networks. regards, Doyle Saylor -------------- next part -------------- An HTML attachment was scrubbed... URL: <../attachments/19981117/2a3116ed/attachment.htm>