Jump to content

Talk:Independent component analysis

Page contents not supported in other languages.
From Wikipedia, the free encyclopedia

Untitled

[edit]

The definition of negentropy you have linked to does not seem relevant to this article; could you update the negentropy article to explain the term as used in ICA literature? Thanks --Chinasaur 00:37, 16 Jun 2005 (UTC)

In literature, approximations to negentropy are not really used to separate signals. Most ICA methods are derived from the maximum likelihood theory, which is equivalent to the infoMAX theory. Both maximize the network entropy, so most methods are approximations to network entropy instead of approximations to negentropy.

See Survey on Independent Component Analysis by Aapo Hyvärinen

For a definition of negentropy see mentioned survey or the paper by

P. Comon, Independent Component Analysis - a new concept?, Signal Processing, 36:287-314, 1994. JasperKlewer 10:27, 21 March 2006 (UTC)[reply]

Not very good

[edit]

This article is really bad. It jumps straight into a discussion of ICA with noise, which is not the normal case considered, but is more related to linear factor analysis. There used to be some good material here but it has disappeared. I am considering offering a complete rewrite of this article. (Tony Bell) —Preceding unsigned comment added by Tony848 (talkcontribs) 02:17, 16 April 2008 (UTC)[reply]

Tony, if you can find the time I think that would be an excellent idea. The article as it stands is sparse (fitting for ICA but not a good thing for articles), the original definition is not to the point, and the mathematical model could be made more clear. Also, the article seems to lack an algorithmic content. It would be useful to outline a few of the more commonly used ICA algorithms. The fact that neither entropy, KL-divergence, or even a formal definition of statistical independence are mentioned is very curious. Also, diagrams would be great. I can take a stab at this fairly soon, but if anyone wants to bootstrap the rewrite, I will be happy to add in figures and overviews of the various algorithms. --Tekhnofiend (talk) 01:21, 22 October 2008 (UTC)[reply]


German article

[edit]

I find the German article on this subject a lot better. Could a native German-speaker have a look? —Preceding unsigned comment added by 130.89.67.57 (talk) 15:00, 12 September 2007 (UTC)[reply]

definition needs cleanup

[edit]

Was the definition copied from another source and taken out-of-context? As a mathematician, I find it incomprehensible. Lavaka 17:33, 16 August 2006 (UTC)[reply]

I am requesting that this article be made less esoteric. Many terms must be taken for granted if the definition is thought to be meaningful. Wilgamesh 21:06, 20 August 2007 (UTC)[reply]

Amen to that. The casual reader would have an extremely difficult time understanding any of this.Beeblbrox (talk) 03:41, 9 February 2008 (UTC)[reply]

Copyviol

[edit]

--F. Cosoleto 16:14, 17 January 2007 (UTC)[reply]

Number of sources

[edit]

The intro said: the ICA methods are not able to extract the actual number of source signals, the order of the source signals, nor the signs or the scales of the sources. But Sepp Hochreiter and Jürgen Schmidhuber showed how to obtain non-linear ICA or source separation as a by-product of regularization (1999). Their method does not require a priori knowledge about the number of independent sources.

So I changed this to: most ICA methods. Algorithms 16:18, 7 June 2007 (UTC)[reply]

Model

[edit]

Why do we have Generative model subtitle with no content? Its immediately followed by Linear noiseless ICA. Someone familiar with ICA details please fill in. --பராசக்தி 21:12, 9 September 2007 (UTC)

Please make this readable

[edit]

This article is unreadable to the average user. Please bear in mind that this is an encyclopedia, not a mathematics journal. Clarification is needed, especially in the lead section, which gives the reader no reasonable basis for understanding the subjects notability, or indeed, purpose. Beeblbrox (talk) 00:50, 13 February 2008 (UTC)[reply]

Well, six months later, the lead section is still not something your casual reader could understand. I would just fix it myself, but I am that average reader and there is just too much specialized jargon in that lead section that I'm not familiar with. Beeblbrox (talk) 16:27, 13 July 2008 (UTC)[reply]

The opening sentences are not good English.The word "supposing" should probably be "assuming". The next two or three sentences are ungrammatical. — Preceding unsigned comment added by AdmiralSven (talkcontribs) 00:20, 16 May 2012 (UTC)[reply]

No time delay?

[edit]

"Usually the problem is simplified by assuming no time delays and echoes."

I'd intuitively think that time delays between each microphone would make it easier to separate the signals, since each source would correlate with itself at different times. If ICA is simplified by removing time delay, does this mean ICA doesn't take advantage of this effect and is not optimal for this scenario? —Preceding unsigned comment added by 71.167.60.246 (talk) 14:30, 1 September 2009 (UTC) Indeed the time difference between the microphones is the base on which the signals are seperated, what the autor means is that the time delays and/(better == ) echoes generated by the room in which the sources are standing are neglected, since these echoes are not statistically independent sources and therefore can not be identified by ICA[reply]

No, the difference in amplitude of the signals at the various microphones is used for separation, the time difference is assumed to be zero. In that sense, although the cocktail party problem is a classical textbook example, ICA is indeed not optimized for this scenario. Also, ICA can not separate more than three sources using only three microphones, which would be possible by an algorithm considering time delays. 129.177.233.236 (talk) 09:08, 13 March 2015 (UTC)[reply]

Unnecessarily complicated algorithm for Generative Model Linear Noiseless ICA

[edit]

Reading through the algorithm, it is apparent that the whole business of inverting the coefficient matrix is completely unnecessary; since you're optimizing the components of the result matrix W directly, just return the result of this optimized matrix instead of returning it as it's inverse/pseudo-inverse A. — Preceding unsigned comment added by 74.243.202.217 (talk) 01:12, 26 July 2014 (UTC)[reply]

[edit]

Hello fellow Wikipedians,

I have just modified one external link on Independent component analysis. Please take a moment to review my edit. If you have any questions, or need the bot to ignore the links, or the page altogether, please visit this simple FaQ for additional information. I made the following changes:

When you have finished reviewing my changes, you may follow the instructions on the template below to fix any issues with the URLs.

This message was posted before February 2018. After February 2018, "External links modified" talk page sections are no longer generated or monitored by InternetArchiveBot. No special action is required regarding these talk page notices, other than regular verification using the archive tool instructions below. Editors have permission to delete these "External links modified" talk page sections if they want to de-clutter talk pages, but see the RfC before doing mass systematic removals. This message is updated dynamically through the template {{source check}} (last update: 5 June 2024).

  • If you have discovered URLs which were erroneously considered dead by the bot, you can report them with this tool.
  • If you found an error with any archives or the URLs themselves, you can fix them with this tool.

Cheers.—InternetArchiveBot (Report bug) 23:38, 12 November 2017 (UTC)[reply]