Preliminary Conference scheduled for lawsuit; imperfect synapses cannot learn

A "Preliminary Conference" has been scheduled for may 21 in the lawsuit brought against me and my son Rafael, which seeks to close the Baiting Hollow Hummingbird Sanctuary, and also demands $3 million in damages. I believe this conference, which I plan to attend, will establish a general timetable for "discovery" by the 2 sides, i.e. the process by which the plaintiffs (the Terrys, the Bherwanis and Ms Hamilton) and the defendants (Paul and Rafael Adams) will seek relevant information (such as website content, emails and other types of exchange). In particular I anticipate that all the parties to the lawsuit, plus some additional parties, will have to make depositions (i.e. formal recorded answers to specific questions posed by one side or the other). The aim of this "Discovery" process is to establish a set of basic facts which all parties can agree on; this would reveal whether the lawsuit has any merit, and what results a trial might lead to, in which case the parties could attempt to settle matters before trial.

Obviously I'm going to have to invest a considerable amount of time and effort into learning about arcane legal procedures. Hopefully I will discover that these procedures are mostly sensible and effective, and designed to efficiently establish true facts and reasonable conclusions, rather than line lawyers' pockets.

It's also possible that the possibility of "mediation" could be raised at this conference. This would involve all parties agreeing to abide by the decisions of an unbiassed external mediator. I've heard rumors that the main requirements of the plaintiffs in such a mediation might be changing the name from "sanctuary" (apparently this name, meant to denote peace, natural preservation and tranquillity, is particularly offensive), and compelling all my visitors to park down near Sound Avenue (eg at the Sound Ave Preserve parking lot) and walk up Terry Farm Road. I would be interested to hear my readers' comments on these issues.

OK, back to science. My colleagues (Kingsley Cox and Anca Radulescu) and I have constructed computer models of networks of "toy" neurons which can learn using Hebbian imperfect synapses.
There are 2 key parts to these models. First and simplest is a way to generate data that can drive learning. One way would be to use a movie, each frame of which is a set of numbers (i.e. pixels). Or one could just use a sequence of "natural" images (e.g. photos of natural scenes, in no particular order. The problem with this is that one does not know what one is supposed to learn from such images - one doesn't really know how they were generated (at least explicitly - of course one might know who directed and filmed the movie etc). And these data are far too complex.

So instead we generate the input data in a completely well-defined way. For example, we can start with a set of random numbers. They are random in 2 senses (1) each number in the set is independently chosen: knowing one number tells one nothing about another (2) they all follow some simple statistical distribution (random means that it can take a range of possible values, each with a defined probability, which constitutes its "distribution". The Gaussian, or bell-shaped distribution is the most familiar example. These random numbers are not directly used as the input: in fact, since they are independent and random, they are completely unstructured - there would be nothing to learn, just as the snowflakes on an old fashioned TV after the channel shuts down are not meaningful. We introduce structure by mixing these random numbers together in a defined way, to generate a new set of numbers. For example, the first number in the new set is created by taking a fixed fraction of the first number in the original, random and independent set, and adding another fixed fraction of the second number in the original set, and so on for all the numbers in the original set.