Talk:Boosting (machine learning)

Bias vs. Variance
The first sentence of the article defines boosting as a method for reducing bias. Isn't this incorrect? If boosting provides generalization, and variance refers to the variance of the model for different training sets (i.e. high variance means overfitting), then boosting should reduce variance and thereby increase bias. I'm confused about this, could someone please comment?

--EmanueleLM (talk) 07:43, 1 June 2016 (UTC) No that's basicallt right since the wrt number of weak learners you can end up with bias (too few of them) or overfitting (too many of them). That's the best paper you can read about Boosting: http://rob.schapire.net/papers/explaining-adaboost.pdf

Anyway boosting almost always reduces bias and in practice unless you use a lot of learners does not increase variance significantly.

Strong vs. Weak
The explanation of strong vs. week learner is a bit confusing. Unfortunately, I am not the right person to explain it better. —Preceding unsigned comment added by 193.171.142.61 (talk) 08:42, 7 December 2010 (UTC)

Boosting
Boosting is also a method for increasing the yield of a fisson bomb (Boosted fission weapon). Is that something that should be linked from this article? Or maybe put on the disambig. page for boost? --81.233.75.23 12:53, 1 June 2006 (UTC)

It should be in the disambug. page. Grokmenow 16:27, 10 July 2007 (UTC)

Oh, didnt see the date. Sorry about that. Grokmenow 16:27, 10 July 2007 (UTC)

Computer vision category
I removed this article from the computer vision category. Boosting is probably used by some people to solve CV problems but


 * 1) It is not a methodology developed within CV or specific to CV
 * 2) Boosting is already listed under the ensemble learning category which is linked to the CV category via maching learning.

--KYN 22:36, 27 July 2007 (UTC)

Recent Articles
I removed two articles from the references section. Perhaps another references section should be started to include some of the additional research on boosting.

-- AaronArvey —Preceding unsigned comment added by AaronArvey (talk • contribs) 01:15, 3 September 2007 (UTC)

"branching program based boosters"
The paper cited in reference to "convex potential boosters [not being able to] withstand random classification noise" states that "branching program based boosters" can withstand noise.

It would be really swell if someone knowledgeable could explain what "branching program based boosters" are. (Sorry that I can't) —Preceding unsigned comment added by 194.103.189.41 (talk) 14:14, 23 March 2011 (UTC)

Agreed! --149.148.237.120 (talk) 09:30, 27 August 2014 (UTC)

Merging article
I think that this article : http://en.wikipedia.org/wiki/Boosting_methods_for_object_categorization

should be merge with this one. Anyone agree ? — Preceding unsigned comment added by 207.139.190.179 (talk) 20:21, 4 December 2012 (UTC)
 * Yes (even if rather belatedly). Klbrain (talk) 14:23, 26 July 2016 (UTC)

Boosting for multi-class categorization
Boosting for multi-class categorization states at the second paragraph that The main flow of the algorithm is similar to the binary case. Perhaps the author intended the word flaw? Anyway there is no mention in the binary case of its main flow or flaw. So this needs to be clarified and possibly rewritten.--Gciriani (talk) 01:57, 3 June 2017 (UTC)

I think he meant algorithm by flow because there is no mention of a flaw. — Preceding unsigned comment added by 2402:4000:2080:1FDC:15CB:7F3F:641:48B5 (talk) 05:28, 23 July 2020 (UTC)

first sentence is contradicted by its own citation
first sentence is:

"Boosting is a machine learning ensemble meta-algorithm for primarily reducing bias, and also variance[1]"

and that citation leads to:

https://web.archive.org/web/20150119081741/http://oz.berkeley.edu/~breiman/arcall96.pdf

I can't find anything in that paper that suggests boosting is "primarily for reducing bias". In fact it seems to be the opposite:

"Although both bagging and arcing[=boosting] reduce bias a bit, their major contribution to accuracy is in the large reduction of variance. Arcing does better than bagging because it does better at variance reduction."

31.220.221.120 (talk) 14:04, 14 December 2017 (UTC)