Talk:Multi-task learning

Hi editors of "Multi-task learning",

As a project for a course I have completed a major edit for this article. Because of the assessment requirements for the course, the edit was completed "offline".

"Multi-task learning" was selected for improvement since it did not contain any mathematical details and much of the text was copy-pasted from various article abstracts and sources around the internet. I attempted to give better attribution, although I left some text unedited. In particular some of Kilian Weinberger's wording remains particularly since I wasn't sure if in fact he was the one who posted it here.

I tried to edit all sections for clarity, relevance, and brevity. I also added mathematical details for RKHSvv taken from Carlo Ciliberto's recent paper, which provides somewhat of an overview of this take on the subject. I suspect the level of detail is too much, as I had the instruction staff of my course as a primary audience in mind.

You can view the newly edited article here. https://en.wikipedia.org/w/index.php?title=Draft:Multi-task_learning&redirect=no Unless there are major objections, I will move the text over here say around 1st Jan.

Thanks Mattheworlando (talk) 20:27, 18 December 2015 (UTC)

External links modified
Hello fellow Wikipedians,

I have just modified 2 external links on Multi-task learning. 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:
 * Added archive https://web.archive.org/web/20041118134329/http://big.cs.uiuc.edu/webpage/cumulativeLearning/cumulativeLearning.html to http://big.cs.uiuc.edu/webpage/cumulativeLearning/cumulativeLearning.html
 * Added archive https://web.archive.org/web/20131224113826/http://klcl.pku.edu.cn/member/sunxu/code.htm to http://klcl.pku.edu.cn/member/sunxu/code.htm

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

Cheers.— InternetArchiveBot  (Report bug) 10:49, 12 January 2018 (UTC)

Needs an Overhaul
I think this page needs a bit of an overhaul/update. It seems oddly focused on mathematical, shallow learning, and robotics applications for a technique that is now used extensively in the machine learning community across a wide range of disciplines (especially deep learning for NLP).

A lot of the existing content should probably move under a "Theoretical underpinnings for why multitask learning works, conceptually" section.

Sebastian Ruder's survey of multitask learning has been quite widely cited so it might be worth using that as a guide for the structure:
 * Types of parameter sharing (hard vs soft vs hierarchical)
 * Theoretical basis for MTL (include much of the existing content here)
 * MTL in non-neural models (include more of the existing shallow-learning discussion here)
 * Deep learning approaches - highlights of DL research, maybe split by field
 * NLP - different task framings (e.g. the T5, decaNLP approach, huggingface's HTML)
 * Vision
 * Etc... — Preceding unsigned comment added by Themoxon (talk • contribs) 13:44, 27 October 2021 (UTC)