Jan 31

Does anyone know anything about Neural Networks?

Archived in the category: neural networks

Any genuine knowledge would be appreciated, not just stuff copied and pasted from wikipedia, I can do that myself. I just want any information anyone might have please.
I should have specified that I am a second year cybernetics student, wishing to expand my knowledge.

Neural network is a programming methodology that uses the puts together different peices of information to generate a pattern. The data collected from different points is weighted (sometimes ) and extrapolation is done using defined rules or from first principals to obtain a better idea of the system without undergoing a tedious test every point approach. Since the system is learning more and more data contributes to better understanding.

My understanding is that the output of neural network program is 2D or 3D mesh graph where in all the known values are used and unknown are extrapolated by an algorithm. For e.g. we could use neural network program to simulate how a model would work under wind velocity, pressure, drag etc and when sufficient points are available extrapolate/simulate the points to arrive at a conditions that are difficult to produce and when an appropriate model fits simulation use it test results in a difficult to produce conditions (for example sending to space). This can be used for any particular circumstances like building a robot, forcasting weather, data mining using existing data etc.

The concept is similar to how brain's learning functions. The term neural networks means connecting different related data through set of rules like neurons collect the pass to the brain.(my way of looking at it)

A standard definition is available in the reference.

powered by Yahoo Answers

admin

5 comments for “Does anyone know anything about Neural Networks?”

1
Dad

its a whole branch of science, basically it is the study of how your brain is wired up… we are trying to find out how so we get computers to wire themselves up so that they can think. So far we have failed to be that clever, no one neural net artificial intelligence (now there is a film to see - "AI" has passed the Turing test yes, after 50 years of research .. no one is even close…

PS Obvioulsy not YOUR brain but in general …
References :

February 1st, 2009 at 5:04 am
2
amansscientiae

They suck.

:-)

No offense. I had exposure to people who worked on them. And they said they suck. I liked my genetic algorithms much better.

If you want more than Wikipedia, get yourself the textbooks in the field. Read articles published at conferences and in journals.
References :

February 1st, 2009 at 5:41 am
3
mavis b

the problem, the brain can 'simultaneously' process billions of bits of information gathered from 6 senses and the compared the results to a data base (memory and experience) and then derive a solution even though it has never encountered the problem.
the second problem, even though there a massive strides to storage, retrieval, and multi-stream processing, the current technology can not overcome the hurdles.
It will take a bio-mechanical break through of multiple impulse lines on a molecular level to overcome the 'walls' to achieve a stable redundant circuit.
References :

February 1st, 2009 at 6:29 am
4
greenshootuk

Neural networks are basically computers which "learn" how to solve a clas of problems. The network (which may be hardware or, more often, simulated in software) comprises a number of nodes usually arranged in layers (or rows). The first row froms the input - one example might be the pixels in an image - the last row would be the output and might be just a single node.

Each node in an intermediate layer takes input from the nodes in the previous layer and generates an output from a weighted function of the inputs. These weights can be adjusted.

The network is "trained" by presenting some test input data and then ,using some clever algorithms, adjusting the weights to give the correct output. A large number of test case are used, adjusting the weights statistically.

Applications are in things like pattern recognition.
References :

February 1st, 2009 at 6:58 am
5
Expt

Neural network is a programming methodology that uses the puts together different peices of information to generate a pattern. The data collected from different points is weighted (sometimes ) and extrapolation is done using defined rules or from first principals to obtain a better idea of the system without undergoing a tedious test every point approach. Since the system is learning more and more data contributes to better understanding.

My understanding is that the output of neural network program is 2D or 3D mesh graph where in all the known values are used and unknown are extrapolated by an algorithm. For e.g. we could use neural network program to simulate how a model would work under wind velocity, pressure, drag etc and when sufficient points are available extrapolate/simulate the points to arrive at a conditions that are difficult to produce and when an appropriate model fits simulation use it test results in a difficult to produce conditions (for example sending to space). This can be used for any particular circumstances like building a robot, forcasting weather, data mining using existing data etc.

The concept is similar to how brain's learning functions. The term neural networks means connecting different related data through set of rules like neurons collect the pass to the brain.(my way of looking at it)

A standard definition is available in the reference.
References :
http://searchnetworking.techtarget.com/sDefinition/0,,sid7_gci212648,00.html

February 1st, 2009 at 7:38 am

Laeave a Reply