Neural Search -- was compression

[markw at mohawksoft.com (markw at mohawksoft.com)]

Actually, it should be pretty easy.

I have the recommendations system which is almost, but not quite, a neural
network engine, and it is pretty darn fast. I have my own search engine
which is also fast, it uses data compression to store all the indexes in
memory (this makes it scale on SMP without being I/O bound).

I have an idea as to how it would work as well.


markw at mohawksoft.com remarked:
|
| Lately, I have been thinking about creating a neural network search
| engine. I used to work for FAST Search and Transfer and am more than
| casually aquainted with the difficulties.

Combinine the two ideas gets a fast neural network search.   What  an
idea! Sounds like an oxymoron.  But I'd bet that people who've worked
on  neural  networks  are  starting  to  figure  out  how  to  do  it
efficiently.   I  haven't  followed  that  literature for some years,
though, so I don't have any feel for it any more.

| I probably have 90% of the code
| already, but chances are someone patented some stupid aspect of it and is
| waiting to sue.
|
| Yup, patents are good for innovation.

Actually, it's set up so that the innovation can still  happen.   You
and I do the development on our own, and show that it works. Then the
big guys with money come in and use the patents to take over from the
innovators and turn it into a product.

"Sell the rights to your stuff to us or we'll bankrupt you  with  the
legal costs." It's an old story.