[MOLOnT]: Human Brain Wiring or The Role of Marihuana in Hum

[Rich Grise]

"Guy Macon" <http://www.guymacon.com> wrote in message
news:rMSdneX-xdVeHjDdRVn-gg@speakeasy.net...

Rich Grise <null@example.net> says...

Would'nt a neural net be almost a natural for OCR? Or is it so obvious
that that's a stupid quiestion?

I worked on a project that used a neural network for optical character
recognition. It was scary how much better it was than other methods.

here is a place to get some code to get started with for those of the
microsoft/.bnet persuasion:
http://www.codeproject.com/dotnet/simple_ocr.asp

And here is one for Linux fans such as myself:
http://sourceforge.net/projects/lwneuralnet/

Cool! I kinda have a project in mind, but while thinking about it
I realized that I should try that OCR S/W that came with the scanner.

The closest to a neural net I'm working on is hacking a couple of
joysticks into a keyboard controller.

Which, while I'm thinking about it, would give a potential ... lessee
- 18 (!) lines to one side of the matrix, 8 to the other, so more
than 128 inputs.

Of which any arbitratrary number could be used to make a jeopardy
board. Or a dartboard sensor.

Cheers!
Rich

 

[Terry Given]

"Guy Macon" <http://www.guymacon.com> wrote in message
news:KvWdnZyZSr4JxDDdRVn-hg@speakeasy.net...

Rich Grise <null@example.net> says...

I like to play headgames with self-appointed netcops.

In my opinion, there is a world of difference between a self-appointed
netcops and a person making a polite request that everyone is free to
ignore. There are many people who have a strong desire to control
others, but there is no reason to assume that every attempt to get a
group of people to agree on a course of action is in that category.

hear hear. self appointed netcops are more like DarkMatter, and seem to
think calling people "fucking retards" is somehow going to help.

Ultimately, I think Guys reasoning is this: As engineers, we voluntarily use
standardised methods every day (we all agree, for example, on prefixes like
M,m,k,p, and we all know what "hysteresis" means). It therefore should be
straightforward with such a group of people to introduce an analagous
standard ([OT]), which like the others makes communicating easier.

Terry

 

[Terry Given]

"John Larkin" <jjlarkin@highlandSNIPtechTHISnologyPLEASE.com> wrote in
message news:ueipa0558eo63s5k38gcikru35pmqpi0u7@4ax.com...

On Thu, 20 May 2004 04:30:07 GMT, "Rich Grise" <null@example.net
wrote:


[snipped huge amounts of priceless stuff]


This also makes me wonder - whatever happened to neural nets?


Met fuzzy logic and ran away together, I hope.

Does anybody know of a neural net application that actually works?

John

depends on your definition of "works." Let "works" = "people pay me to write
papers about this crap" in which case the existence of uni. professors
writing such papers proves it "works."

I have seen many pointless smps papers, which should be titled "A dreadfully
expensive and time-consuming neural-net approach to greatly over-complicate
the job of smps loop control, requiring ludicrous amounts of computational
power for a task which an LM324 can do nicely for $0.10"

Terry

 

[John Popelish]

Terry Given wrote:
(snip)

I have seen many pointless smps papers, which should be titled "A dreadfully
expensive and time-consuming neural-net approach to greatly over-complicate
the job of smps loop control, requiring ludicrous amounts of computational
power for a task which an LM324 can do nicely for $0.10"

Amen, brother.

--
John Popelish

 

[Scott Stephens]

Terry Given wrote:

I have seen many pointless smps papers, which should be titled "A dreadfully
expensive and time-consuming neural-net approach to greatly over-complicate
the job of smps loop control, requiring ludicrous amounts of computational
power for a task which an LM324 can do nicely for $0.10"

Bravo!

--
Scott

**********************************

DIY Piezo-Gyro, PCB Drill Bot & More Soon!

http://home.comcast.net/~scottxs/

**********************************

 

[Terry Given]

"Scott Stephens" <scottxs@comcast.net> wrote in message
news:bNWrc.8363$hi6.912369@attbi_s53...

Terry Given wrote:

I have seen many pointless smps papers, which should be titled "A
dreadfully
expensive and time-consuming neural-net approach to greatly
over-complicate
the job of smps loop control, requiring ludicrous amounts of
computational
power for a task which an LM324 can do nicely for $0.10"

Bravo!

--
Scott

Here is where I also recognise that real applications do exist for this
technology, as others have mentioned (image recognition etc). I just think
its stupid to throw 20,000,000 transistors at a job when 30 will do. hell,
if 2 sot23's can do the job and are cheaper, i'll use them.

A big downside of microprocessors is that they need code. It is sometimes
difficult to persuade managers that products arent finished until both the
hardware AND the software function correctly under the desired operating
conditions. Once an analogue circuit design is operational, the job is
done - providing of course you make meaningful measurements to prove it does
work, and that your design accounts for variations like temperature, time,
tolerance, vendor, pressure, contamination, etc. But that has to be done for
the microprocessor hardware anyway...

Terry

 

[Will Dwinnell]

John Larkin <jjlarkin@highlandSNIPtechTHISnologyPLEASE.com> wrote:
"Does anybody know of a neural net application that actually works?"

Sure, I've built loads of them for paying clients over the years.
Examples include: predicting machined metal part quality based on
process parameters, industry revenue forecasting and account-level
financial customer attrition prediction based on customer attributes
and behavior, to name a few.

-Will Dwinnell
http://will.dwinnell.com

 

[Sir Charles W. Shults III]

"John Larkin" <jjlarkin@highlandSNIPtechTHISnologyPLEASE.com> wrote in
message news:ueipa0558eo63s5k38gcikru35pmqpi0u7@4ax.com...

Does anybody know of a neural net application that actually works?

It seems that they used neural nets to learn handwriting for the Post
Office machinery, but I might be mistaken. I know that it has been used for
controlling temperatures in large buildings pretty effectively, where you
have lots of air ducts with butterfly valves and two or three HVAC units.
In that case, the net is simulated in software (which is pretty much the
best way to go about it) and learns to control the airflow and get the most
effective energy management out of it.
I also have heard that some companies in the speech synthesis and
recognition game have used neural nets to learn the bacis ground rules and
then refine their algorithms. For the most part, neural nets have fallen
out of favor due to the unpredictable nature and the often difficult to
debug problems. But they certainly have their uses in random learning
environments.

Cheers!

Sir Charles W. Shults III

 

[Rich Grise]

"Sir Charles W. Shults III" <NOaichipSPAM@cfl.rr.com> wrote in message

"John Larkin" <jjlarkin@highlandSNIPtechTHISnologyPLEASE.com> wrote in
Does anybody know of a neural net application that actually works?
I also have heard that some companies in the speech synthesis and
recognition game have used neural nets to learn the bacis ground rules and
then refine their algorithms. For the most part, neural nets have fallen
out of favor due to the unpredictable nature and the often difficult to
debug problems. But they certainly have their uses in random learning
environments.

I was doing a little project once, with the intent of doing speech
input. I had noticed on some educational program, that they took
newborn babies, and could train them to react on certain vowel
sounds, and the babies would react the same to the same vowel,
regardless of the pitch or timbre of the speaker's voice. The
vowel has some recognizable pattern that's independent of the
fundamental. I'd heard of "formant filters," so I thought I'd
look at the spectrum. I built 8 BPFs between about 250Hz-3KHz,
and displayed 8 bar graphs on my terminal. And it showed the
spectrum of your voice. And no matter who was talking, each
vowel sound (or probably more like each phoneme) has a spectral
pattern of its own.

So I was kinda looking into how to do pattern recognition,
i.e., if the graph is a certain shape, it resolves to a particular
phoneme. But I got bogged down, and moved on to different things,
and one thing after another, that's about where it sat. It sounds
like something that a neural net would be a natural for. But I'd
have to go to school for awhile to try to design anything like
that!

Cheers!
Rich