Talk:Feed-forward

From Wikipedia, the free encyclopedia

WikiProject Systems
This article is within the scope of WikiProject Systems, which collaborates on articles related to Systems science.
Systems rating: Start Class High importance  Field: Control theory
Please update this rating as the article progresses, or if the rating is inaccurate. Please also add comments to suggest improvements to the article.
Image:Neuroethologylogo.gif
This article is within the scope of the Neuroethology WikiProject, a collaborative effort to improve Wikipedia's coverage of Neuroethology. If you would like to participate, you can visit the project page, where you can join the project and see a list of open tasks.

[edit] Discussion 2004

The present form of the article describes the term "feed-forward" in a confusing way. At least as far as uses in aspects of "control theory" are concerned "feed-forward" and "open loop" have more in common than the corresponding current Wikipedia articles teach. It would be great if some experts voiced their thoughts and could share them with the public - thanks.

My first thought on seeing this page was that it would be best moved to Feed-forward neural network or some such, but looking at what links to it, it seems you are intending more of a page about the term "feed-forward" in general. My worry in that case is that you'll find yourself with not enough to write, and will be informed that Wikipedia is not a dictionary. I guess that depends what other fields use the term: perhaps it could end up as more of a disambiguation page - IMSoP 01:44, 4 Apr 2004 (UTC) (I'm not following you around, honest, I just fired up recentchanges and saw something I recognised)


Eh, physiology? Is this some new kind of "living perceptron"? Oh well, I guess I'll stop interrupting you and assume you know what you're doing. The reason I put "refers to" rather than "is", by the way, is because "feed-forward" isn't a noun - or I've never seen it used as one, anyway; so if you wanted to use is, you'd have to have a noun for the adjective feed-forward to modify: "...a feed-forward network is one which..." - IMSoP 02:05, 4 Apr 2004 (UTC)

Firstly, thanks. I like it when people review what I enter.

Secondly, I think you may be on the right track (I will move this page to "Feed-foreward regulatory network". Bensaccount 02:08, 4 Apr 2004 (UTC)

I guess I was trying to merge OneLook dictionary and textbook information in an attempt to figure out what "feed-forward" was. Bensaccount 02:19, 4 Apr 2004 (UTC)

Well, I can only comment on what I know, but in Artificial Intelligence, a feed-forward network would be an artificial neural network - such as a simple multi-layer perceptron - which had multiple layers connected in series, such that no neuron received input from any neuron to which its own output contributed. In other words, one in which there is no feedback from later processing to influence earlier layers. Such architectures allow more complex processing than single-layer networks, but are more easily analysable than more complex architectures such as fully-connected networks.
I won't copy that onto the page just yet, because I'm not entirely sure whether that was or wasn't the sense of the term you were originally aiming to explain, but I think you have become confused if you are associating perceptrons with physiology. - IMSoP 19:18, 4 Apr 2004 (UTC)

Feed-forward seems to be used in two contexts: All I know is that in the physiology context, feedforward is a type of neural regulatory system of the nervous system. I dont know the computing meaning. I seem to have mixed the two meaning up because I didn't realize they were different things. If you can fix it you should, I will be back later.Bensaccount 20:04, 4 Apr 2004 (UTC)



Quoting from the entry:

When a hill is encountered the car slows down below the set speed. This speed error causes the engine throttle to be opened further, bringing the car back almost to its original speed. Almost its original speed because a feedback system needs some residual error that can be multiplied by loop gain to provide the necessary correction factor for the duration of the hill.

This is incorrect, but I'm not sure how to succinctly explain what is wrong. I'll go on about it here, instead.  :-)

What the article says is partly true: for a feedback control system which has only a proportional term, there will always be a residual error which the proportional term is unable to resolve.

In the early days of closed-loop control (in factories, say of a chemical process), people first discovered you could regulate the output rate of a process by measuring the deviation from a desired rate (set-point) and then altering the input flows by an amount proportional to that deviation.

The constant of proportionality was determined empirically. Immediately, folks noticed that the output of the system never quite reached the set point, but always ended up a bit below it. The operators quickly realized that they could measure this steady-state error and nudge the input controls up a bit by hand, bringing the system up to the exact desired set point value. As the system approaches the set point, the error drops to zero and so this so-called "integral term" in the feedback equation ceases to affect the systems behavior.

The operators were doing something akin to integrating the residual error over time, and adding that (times a constant of proportionality) back to the input rates. Soon, people realized that this too could be done automatically.

And the car's cruise control, being just such a system, does the same thing. It doesn't need any residual error to maintain the velocity; it simply compares the velocity to the set point and alters the throttle setting according to the control law of the system. See also "PID loop control".

Agreed, the page is incorrect, I was quite shocked to read something so wrong on the Wiki, I'll try to fudge it, but I don't really get what the point the person writing it was trying to make.

Hello chaps. This is a tricky one. As a cybernetician a strict distinction between feedforward and feedback to me has always seemed elusive. Just because an envirionmemtal parameter is monitored doesn't necessarily imply feedforward. If the controlling model is seen as forecasting it still need not imply feedforward. Anyway this graphic may help. It's redrawn after Fig 1-1 in Mrosovsky's "Rheostasis" OUP 1990. It's defintely worth discussing!

Image:FeedForward.png

Incidentally someone ought to discuss homeostasis or homeodynamics as Yates calls it, in one of these Wiki entries on Feedback. Stasis seems incorrect for an essentially dynamic process even though convergence to a fixed point is implied. Mrosovsky's fixed point moves hence his title but whither the much loved Rheostat of the electrical laboratory? This probably a bit off topic and should be raised in the Homeostasis entry but maybe it illuminates some of the difficulties.--Nick Green 17:08, 2 September 2006 (UTC)