Jump to content

Talk:Unsupervised learning

Page contents not supported in other languages.
From Wikipedia, the free encyclopedia
This is an old revision of this page, as edited by Hyeonjungko (talk | contribs) at 16:54, 19 October 2018 (Update Advanced Writing in the Tech Professions assignment details). The present address (URL) is a permanent link to this revision, which may differ significantly from the current revision.

Template:Findsourcesnotice Template:IEP assignment


Expand

Sentences like "Another form of unsupervised learning is clustering, which is sometimes not probabilistic" and "Also see formal concept analysis." need to be expanded. -130.49.221.7 14:41, 6 February 2006 (UTC)[reply]

Agreed. In addition, the entire article as of today is very badly written. Robbyjo 21:23, 18 January 2007 (UTC)[reply]

In my eyes, this article is very misleading. K-Means or spectral clustering algorithms are some of the most prominent unsupervised learning algorithms. All of them assume data a priori.

This article is misleading and it also assumes a lot of prior knowledge with out explaining its terms very well. Are there other tags that can be added to this article to warn future readers?--Stewartm82 (talk) 13:31, 9 May 2010 (UTC)[reply]

Cost functions

From Neural network I gather that the cost function is different than for Supervised learning, but this article does not explain the issue. Is it like in compression: we don't care what the output would be (the compressed sequence), but we have a cost = the mean length of compressed messages? CDaMama 15:24, 15 April 2007 (UTC)[reply]

Reinforcement Learning

Perhaps the most widely used type of unsupervised learning (at present) is reinforcement learning, but there's nothing on it here. Needs at least a link to the RL page. --72.224.113.64 12:31, 16 August 2007 (UTC)[reply]

What does "examples" mean?

On this page, and the main machine learning page, the phrase "labeled examples" is not explained or defined before being used to describe things. Can somebody come up with a concise definition? --Bcjordan (talk) 16:30, 15 September 2009 (UTC)[reply]

Does a neural network really belong to unsupervised learning?

Since a neural network has to be trained with preclassified training data I assume this belongs to supervised learning --Ormium (talk) 09:21, 1 February 2011 (UTC)[reply]

No, not all neural nets have training and testing phases. unsupervised neural networks do unsupervised clustering.

Indeed, Hebbian learning is the classical example of unsupervised learning, and stems from work on neural networks (Donald Hebb). MNegrello (talk) 19:58, 11 May 2016 (UTC)[reply]

Robotics attention needed

  • Address issues on article banners
  • Update and expand

Chaosdruid (talk) 10:57, 16 March 2011 (UTC)[reply]

What does "unlabeled data" mean?

What does "unlabeled data", as used in the article's opening sentence, mean? Duoduoduo (talk) 22:01, 4 February 2013 (UTC)[reply]

From what I understand, it means there is no ground truth. The expected class (or label) output is unknown for all data. — Preceding unsigned comment added by 184.161.104.234 (talk) 22:53, 4 July 2014 (UTC)[reply]

Precisely. Labeling the data is the essence of supervision. MNegrello (talk) 20:00, 11 May 2016 (UTC)[reply]

Pick Finder

Is it a kind of 1D Unsupervised learning? If you think about a possible algorithm. It has to exclude noise. Which part is "supervised" and which part is not? Given noise? — Preceding unsigned comment added by 84.75.169.43 (talk) 19:55, 18 July 2015 (UTC)[reply]

Extended Examples

Hello,

My name is Mary-Kate Whelehan and in the interest of transparency, I am an employee at Darktrace - the cyber security company listed in this article as an example. A few details regarding my company in this article are incorrect - I have listed them below in the hopes that the editors may consider these changes.

The assertion that “the system is not fully self-contained, and a team of analysts in Cambridge, UK review results in order to create security alerts” is inaccurate. In fact, the Darktrace system is fully self-contained: it is delivered as an appliance, which is deployed at the core of the network, and uses unsupervised machine learning to detect abnormal behaviors autonomously, without the use of training data or manual input. (Indeed, the system takes just one hour to be installed and produces results immediately, without tuning or configuration) It is correct to state that Darktrace has a team of analysts in Cambridge, UK, although it should be noted that they are also based across the United States and Asia. It is however incorrect that this team “create security alerts”The Darktrace self-learning system automatically generates security alerts. Many customers also use the services of Darktrace’s analysts for the production of in-depth intelligence reports that document cyber-incidents and assist with investigations.

I have cited an article here that may help clarify some of the above details. [1]

Please let me know if there are additional steps I need to take.

Thank you in advance!

Best, Mkwhelehan (talk) 03:51, 9 January 2017 (UTC)MK Whelehan[reply]


This whole section is basically a long ad for darktrace — Preceding unsigned comment added by 31.168.227.138 (talk) 15:03, 10 August 2017 (UTC)[reply]

Disclosure before editing

I am about to make a couple of small factual changes to this page, which I believe are helpful and will fall under the category of non-controversial edits. The reason I mention this at all is in respect for the Wikipedia guidelines re: Conflict of Interest Editing. As I have mentioned above, I am an employee of Darktrace and would like to make a few factual edits to the use of our company as an example. Please don't hesitate to reach out to me at my discussion page if you have any questions. Many thanks.Mkwhelehan (talk) 18:43, 17 January 2017 (UTC)MK[reply]

  1. ^ Hall, Susan. "Darktrace Automates Network Security Through Machine Learning". The New Stack. 5 Jan. 2017 9:58 AM