Archive for the ‘Machine Learning’ Category

Compression Methods

Thursday, October 30th, 2008

I count 3 ways to compress data:

  1. Use short strings/symbols often, long strings rarely (e.g. Huffman coding, Zipf’s law’s effect on words, I me you the / prestidigitation onomatopoeia)
  2. Refer back to things (e.g. zip, gif, lz, “One if by land, two if by sea.”, symbols)
  3. Remove unwanted stuff (e.g. jpg, mpg, stop reading boring stuff)

Are there any others?

Harry Potter – the new Star Trek

Thursday, July 26th, 2007

It occurs to me that Harry Potter will drive a future generation’s idea of where tech should go. Kinda like Star Trek has driven so much over the last 40 years.

Consider spells. They are rather like the search engine query interface. A few simple words and … magic.

But consider the problem from the spell’s point of view. The spell needs quite a lot of processing power. If your spell is to “freeze” something, do you mean the virus 1 meter in front of you? Or, do you mean your friend very near the general direction your wand is pointing? What’s the spell’s target and intention?

And consider the wand. Its purpose, apparently, is the help the spell figure out your intention. Is a wand the best thing to use?

Pointing works very, very well to indicate many things. But what if the “thing” is not a physical thing in space? What if you want to freeze a discussion? “Everyone stop talking for a moment” (everyone being, presumably, slaves – robots – machines). Let’s say you have a half dozen home-building machines sawing and hammering away on the new house. Pointing the wand and saying “freezzzaaam” is really, really ambiguous. Maybe you mean that the place in the house you’re pointing at is “just right”. But maybe you want all of the robots to stop working and take a lunch break or something. Or maybe they are all sorta working at cross purposes and need to stop and take a breath. Or maybe you’re putting another robot in the group and the others need to stop for a moment to regroup.

If things work out the way they apparently will, such things will be important problems to solve.

Consider, for instance, this simple example: Garbage trucks.

How do they work? Let’s assume low tech here. Nothing fancy.

The garbage guy is in a nice, comfortable cab, monitoring what’s happening with his truck. Maybe he takes the wheel in locations that are not handled by the auto-driver – like running through town to the freeway back to the dump, for instance. But, the truck does all the work while slowly prowling up and down the residential streets, flipping garbage cans’ contents in to the truck. The arm that grabs the cans can see the cans, grab ‘em, empty ‘em, and put ‘em back on the curb. Not a big deal. Especially late at night or early in the morning. Slowly driving the truck down these streets really amounts to dodging any kids there may be. Late night, early morning hours makes that job pretty easy, too.

So the operator, the garbage man, is a monitor, a watchman. He may occasionally need to get out to unjam something. But on the other hand, he’s more like the brakeman on a train, isn’t he?

Which puts him back at the shop rather than in the cab. He’s monitoring a dozen trucks. When he has a problem with more than one at one time he simply stops the others while each problem is dealt with. Freezing a bunch of trucks is troublesome in traffic but not on the garbage-can streets.

So, when the trucks need to go in traffic, do they flock together?

Anyway, what’s this guy in the shop going to point his wand at?

More Compression

Saturday, June 24th, 2006

Speaking of compression: I’m having a hard to time distinguishing “compression” and “establishing meaning” in the context of the thinking process.

More Expected Characters

Friday, June 9th, 2006

Now, it’s expected words.

Or, more exactly, after running the Buffet letters through a program that tracks strings of words (rather than characters), the last of a sequence of letters is shown with the words that are in common strings made small. And unusual words or strings of words are made big.

Common strings are small - uncommon strings are big

The effect is the same. Boiler plate paragraphs are small. New stuff is big.

Data Compression

Saturday, May 6th, 2006

I count three ways to compress data:

  1. Make common quantas of data short, uncommon ones long. e.g. Huffman encoding. I, am, not, be, a, or, prestidigitation, gesticulate, onomatopoeia, redundant.
  2. Reference known data. e.g. Symbols. ZIP file encoding of references to repeated byte strings. Refer to a whole book’s worth of information by referencing the title. One if by land, two if by sea.
  3. Drop information that is not needed. e.g. JPG images. MP3 music. Forget it all. Don’t do it.

Are there any more?

In a sense, all optimization is data compression, is it not?

How many MIPS does the Earth’s DNA crank

Monday, March 6th, 2006

I’ve often wondered how much processing power the Earth’s DNA has. In a sense, genetic evolution seems to be a search function, not unlike a nervous system, always seeking to represent, in some transformed way, a match to its outside reality.

Hmmm. If there were some reasonable way to measure “MIPS” in a broad sense, then I imagine that some of the parallels ‘tween nervous systems and genetic evolution would be clearer than they are now.

Anyway, thinking of genetic evolution as being a “brain” brings up the question: Is it self-aware?

Which brings up the question of: What the heck does “self-aware” mean?

Or, another question: Does it, genetic evolution, that is, have intention? (I’m thinking that anything that is “self-aware”, whatever that means, probably has motives of some sort. Maybe not true, of course, but it sure feels ok to think so.)

Etc.

It’s the “etc” part that’s fun.