[ml] Fwd: Forgery Detection, Overfitting and Recent Results

Brian Morris cymraegish at gmail.com
Tue Mar 1 20:19:19 PST 2011


handwriting recognition sounds like a good image analysis task for practice
since it is 1-bit (black and white), should be low resource usage. I wonder
what is challenging, it is not a simple clustering discrimination problem ?
Anybody familiar with the science of this (what does a distribution of
samples look like).


by the way I have obtained a face recognition software, looks like a
prototype but complete w/ a good size test data set, I haven't tried it yet
but maybe someone wants to do that together ?
(don't ask me how it works because I haven't yet looked at it).




On Tue, Mar 1, 2011 at 12:25 PM, Mike Schachter <mike at mindmech.com> wrote:

>
>
> ---------- Forwarded message ----------
> From: Anthony Goldbloom | Kaggle <anthony.goldbloom at kaggle.com>
> Date: Tue, Mar 1, 2011 at 12:04 PM
> Subject: Forgery Detection, Overfitting and Recent Results
> To: mike at mindmech.com
>
>
>   [image: Kaggle Update]
>   *Tuesday, 1st March 2011*
>   New Competitions
>
>  For those looking for some practice before the $3 million Heritage Health
> Prize, we've just launched two new competitions.
>
>  The first requires participants to develop an algorithm that can identify
> handwriting<http://kaggle.us1.list-manage.com/track/click?u=e4c8fb8b43860678deab268e5&id=51983b1ef8&e=bf02f945c6>.
> Such algorithms are important in the forensics world, where they are used to
> detect forgery. The competition is associated with the International
> Conference on Document Analysis and Recognition<http://kaggle.us1.list-manage2.com/track/click?u=e4c8fb8b43860678deab268e5&id=b334ae28ef&e=bf02f945c6>,
> to be held in Beijing in September.
>
> The second is hosted by Phil Brierley, winner of the tourism 2 competition
> and author of the Tiberius<http://kaggle.us1.list-manage.com/track/click?u=e4c8fb8b43860678deab268e5&id=7316cca577&e=bf02f945c6>predictive modeling software. Phil has noticed that it's often the case that
> the person who is leading on the public leaderboard, doesn't win because
> they have overfit their model. Phil has setup a competition with a
> simulated dataset<http://kaggle.us1.list-manage1.com/track/click?u=e4c8fb8b43860678deab268e5&id=1a7db52a41&e=bf02f945c6>, particularly
> designed to give participants practice at not overfitting.
>
>  Recent Results
>
> Since our last newsletter, the RTA competition and the Melbourne University
> competition have closed.
>
>  The RTA competition<http://kaggle.us1.list-manage1.com/track/click?u=e4c8fb8b43860678deab268e5&id=29cef9c92b&e=bf02f945c6>,
> requiring participants to predict travel time on Sydney's M4 freeway, was
> our biggest competition so far, attracting 364 teams. The winners were José
> Pablo González<http://kaggle.us1.list-manage1.com/track/click?u=e4c8fb8b43860678deab268e5&id=b282ffce51&e=bf02f945c6>,
> a PhD student at Carnegie Melon University and Guido Matias Cortes<http://kaggle.us1.list-manage1.com/track/click?u=e4c8fb8b43860678deab268e5&id=67f2f73d17&e=bf02f945c6>,
> a PhD student at the University of British Columbia. Congratulations also to
> Sergey Yurgenson and Alisson Azzolini who finished second and third. There's
> a nice writeup of the competition on ZDNet<http://kaggle.us1.list-manage1.com/track/click?u=e4c8fb8b43860678deab268e5&id=95392d1900&e=bf02f945c6>
> .
>
>  The Melbourne University competition<http://kaggle.us1.list-manage.com/track/click?u=e4c8fb8b43860678deab268e5&id=27a4fbc35f&e=bf02f945c6>,
> which had participants predict which grant applications would be successful,
> was won by Kaggle's own Jeremy Howard (he entered before he joined Kaggle).
> However, he was ineligible, so Quan Sun<http://kaggle.us1.list-manage.com/track/click?u=e4c8fb8b43860678deab268e5&id=c12f39d848&e=bf02f945c6>,
> a PhD student at Waikato University in New Zealand collected the prize.
> Third placed went to Yuanchen He, a senior engineer at McAfee. Each has
> posted their methodogy on the Kaggle blog (Jeremy<http://kaggle.us1.list-manage.com/track/click?u=e4c8fb8b43860678deab268e5&id=7266074157&e=bf02f945c6>,
> Quan
> <http://kaggle.us1.list-manage1.com/track/click?u=e4c8fb8b43860678deab268e5&id=f2c792557c&e=bf02f945c6>and
> Yuanchen<http://kaggle.us1.list-manage.com/track/click?u=e4c8fb8b43860678deab268e5&id=9356a7b8d7&e=bf02f945c6>
> ).
>    Visit Kaggle<http://kaggle.us1.list-manage1.com/track/click?u=e4c8fb8b43860678deab268e5&id=68bd8eeb0a&e=bf02f945c6>today, or discover more about Kaggle and read interesting articles at No
> Free Hunch<http://kaggle.us1.list-manage1.com/track/click?u=e4c8fb8b43860678deab268e5&id=35f6665e21&e=bf02f945c6>,
> our regularly-updated blog.
>   [image: Kaggle]<http://kaggle.us1.list-manage.com/track/click?u=e4c8fb8b43860678deab268e5&id=e37fe72458&e=bf02f945c6> [image:
> LinkedIn]<http://kaggle.us1.list-manage.com/track/click?u=e4c8fb8b43860678deab268e5&id=0c955828b9&e=bf02f945c6> [image:
> Facebook]<http://kaggle.us1.list-manage.com/track/click?u=e4c8fb8b43860678deab268e5&id=3d73eb5f76&e=bf02f945c6> [image:
> Twitter]<http://kaggle.us1.list-manage1.com/track/click?u=e4c8fb8b43860678deab268e5&id=31a568ddfb&e=bf02f945c6>
>
> View in browser<http://us1.campaign-archive2.com/?u=e4c8fb8b43860678deab268e5&id=234ef32a6e&e=bf02f945c6>•
> Unsubscribe<http://kaggle.us1.list-manage2.com/unsubscribe?u=e4c8fb8b43860678deab268e5&id=f42f9df1e1&e=bf02f945c6&c=234ef32a6e>• Forward
> to a friend<http://us1.forward-to-friend.com/forward?u=e4c8fb8b43860678deab268e5&id=234ef32a6e&e=bf02f945c6>
>
> You are receiving this email because you are a Kaggle user or you are on
> Kaggle's mailing list
>
> Copyright 2011 Kaggle All rights reserved.
>
>
>
> _______________________________________________
> ml mailing list
> ml at lists.noisebridge.net
> https://www.noisebridge.net/mailman/listinfo/ml
>
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: http://www.noisebridge.net/pipermail/ml/attachments/20110301/09e19c2e/attachment-0001.htm 


More information about the ml mailing list