Forgot your password?
typodupeerror

Comment: There's already a textbook (Score 1) 119

by ulatekh (#47903365) Attached to: The Challenges and Threats of Automated Lip Reading

The most obvious approach is to combine the 2 methods - much like humans do, especially in noisy environments.

Obvious, indeed. There's already a textbook for the subject, Multimodal Signal Processing...available for free online, no less.

This is exactly the sort of system you'd want on a flight deck, to supplement the accuracy of speech-recognition in the presence of noise, especially intermittent noise such as turbulence. It can also help with speaker identification.

As for the hopelessly naive idea that "society" should be able to choose whether this sort of thing should exist...the textbook came out in 2009.

Comment: Denoise them, or don't bother (Score 1) 130

by ulatekh (#47858979) Attached to: Ask Slashdot: Best Service To Digitize VHS Home Movies?

Years ago, I became obsessive with producing high-quality DVDs from my extensive VHS/LaserDisc collection. Eventually, that led to the creation of y4mdenoise, part of the mjpegtools package. If you're willing to spend the time to let your computer chew on your digitized video, this tool will squeeze virtually all of the noise out of your signal.

Without it, you're wasting most of your bitrate just to encode noise. A video encoder can't tell the difference between noise and high-frequency detail.

If you don't want to spend that much time, then yuvdenoise, also in the mjpegtools package, does pretty well too.

Comment: Damn it, hire hackers as security professionals! (Score 1) 111

by ulatekh (#47858881) Attached to: Home Depot Confirms Breach of Its Payment Systems

Yet another major computer security breach at a big retailer, compromising the payment details of uncountable customers.

It seems to me that the core problem is that companies won't hire actual experienced hackers as security consultants; for some reason, the idea terrifies them. Instead, they hire bozos that possess some worthless "security" certificate (like CompTIA).

Or even worse, they'll hire a hacker that was dumb enough to get caught and go to jail for his actions. For some reason, that gives them credibility.

Those of us who managed to spend their teenage years hacking everything in sight, and not getting caught — the ones with real expertise — get nothing.

And so these breaches continue.

Oh, and BTW, this is why I pay cash.

Comment: The limit is human (Score 1) 168

by ulatekh (#47686937) Attached to: Processors and the Limits of Physics

Your reasoning is false. Most AI algorithms are having a high level of parallelism which make them less susceptible to the single CPU physical limit. You can achieve incredible performance improvement on GPU and other parallel architectures.

Good luck finding enough programmers that can write code with that level of parallelism.

Most of the multithreaded code I encounter in the real world simply slaps mutexes around things, whether or not they're needed, or even applied consistently. Most of the time, the mutex could be replaced with something cheaper, like atomic operations, or even unique state-transitions on a single volatile global variable.

Your experience may differ. Maybe I just have the back luck of working with morons most of the time.

Comment: Marty, you're not thinking 4th-dimensionally! (Score 1) 109

by ulatekh (#47686919) Attached to: Why the Universe Didn't Become a Black Hole

Perhaps the answer doesn't lie in the 3rd dimension.

One of the possible consequences of the curvature of 4th-dimensional space-time is that our universe may be a 3-dimensional surface of a 4th-dimensional hypersphere. And if the 4-dimensional universe is expanding, the 3-dimensional universe would expand too.

This model of the universe was also used in a famous sci-fi novel.

Comment: Programming general intelligence (Score 1) 161

by ulatekh (#47660223) Attached to: New Watson-Style AI Called Viv Seeks To Be the First 'Global Brain'

No one knows how to program general intelligence.

Well, I have an idea on how to crack that problem...but I'll never have the time and energy to pursue it. I'm also a terrible salesman, so I'll never convince anyone to fund it.

The first part involves defining the goal properly. What's the point of making a computer that's intelligent like a human being? A computer is not a human being. If one wants to make an intelligent computer, it must be done in a way that makes sense given the nature of a computer. There's a difference between artificial intelligence (e.g. what you put into video games to make NPCs interesting) and machine intelligence (e.g. what you put into a jet fighter so that it creams the enemy). Most efforts I see seem to revolve around achieving the former.

It would require a programming language that essentially allows new statements to be added to the language as easily as most OOP languages allow a subclass to be written. The general format of the language would be human-readable text, e.g. English. You don't start off by trying to get it to understand silly world problems, like the word "respectively" — that's a relatively sophisticated ability that comes much later. You just get it to understand the world it can see (i.e. the parts of a computer and its peripherals), with the definitions tracing back to the one concept it can understand — "I". After a fair bit of hand work, you'll have a system that can read normal human text and write code to consolidate its understanding of what it read. Imagine a natural-language parser on the front end and something like llvm's cross-platform assembly-language on the back end.

Once it's able to learn some basic knowledge, the first priority should be to teach it how to program a computer. When it gets to the point that it understands enough about computer programming to reflect upon its own implementation, then it can take over its own development, and then it starts growing exponentially.

There's a lot more to my plan — I've had it for "some time" — but there's no point in spilling all the beans at once.

I don't know if anyone out there has ever tried to design a machine-intelligence along these lines, but I've never heard of one. I'd be interested in hearing about any existing work in this direction.

Say "twenty-three-skiddoo" to logout.

Working...