41

I am currently writing my bachelors thesis about recognition of mathematical symbols. This is supposed to be followed by a masters thesis about recognition of mathematical formulas. (I already spoke with the Professor; it was his idea to split this task.)

So the practical part of my bachelors thesis will be very similar to Detexify. Except that I will use much more sophisticated classification methods to (hopefully) get much better classification accuracy and higher speed than what Detexify provides. (Detexify uses Dynamic Time Warping; I will use Artificial Time Delay Neural Nets, probably coupled with Hidden Markov Models.)

But there is one problem: Those machine learning tools need much data. My advisor told me that I should have at least 100 training instances per symbol I want to recognize.

The current state is here: LaTeX symbols by number of training instances

So I need people that go to http://write-math.com/train, click on a challenge or a symbol and draw that symbol.

In case you have an idea how to structure the interface better, just let me know. I will try to improve that.

I've informed fellow students about my thesis in lectures, I twittered, Google+, Facebook, E-Mail lists, Reddit ... I think I tried everything. Except getting into contact with this community.

The question is: What is the right way to get into contact with people of this community?

I only have a very vague idea what CTAN and TUG are. Are there probably email lists I could contact?

Update

I've got about 2,000 datasets in the last two days. That's amazing! Thank you very much. But I still need A LOT more. At the moment, only 3/~180 symbols have the necessary 100 training instances (see overview).

To give you a little insight in some of the work I do, I've just create a page where you can see how some pre-processing algorithms change the data:

Playing with pre-processing

enter image description here

17
  • 1
    It is the right way to contact with me. :-) I hope that you question will not be closed.
    – Przemysław Scherwentke
    Commented Jun 19, 2014 at 4:38
  • 3
    While I admire the gesture and sentiment, this is not so much about LaTeX as it is about an experiment. Your posed question makes this post seem more fitting on TeX - LaTeX Meta, but then it's not really about this site. It's more about the community. You're right, this post may not have a definitive answer here, making it a mixture of being too broad and off-topic. What about writing a blog post? Why not visit the chat room? Ask people there whether they could spread the word, or point your to blogs that you can write to...
    – Werner Mod
    Commented Jun 19, 2014 at 5:40
  • what is your timeframe? a message to [email protected] and a posting on comp.text.tex would reach a somewhat overlapping audience. an announcement in tugboat isn't out of the question, but it wouldn't be published until late fall; send your request to [email protected] and the editors will get in touch with you (but they might also ask you to write a report after the fact ;).
    – barbara beeton
    Commented Jun 19, 2014 at 13:00
  • @barbarabeeton: I should finish the thesis in September, so I should have the data to run the experiments on until end of August. It would be nice to get data after that, but that data would not get into the thesis. However, other people could use it. (And I really think that the finished result will be better than Detexify so I am still interested in making the website well-known after my bachelors thesis.)
    – Martin Thoma
    Commented Jun 19, 2014 at 13:47
  • 2
    @moose: We have the TeX.SE Blog. In chat, talk to our most talkative Paulo; he'll be glad to help set you up with connections. Currency of payment is usually in ducks... ;)
    – Werner Mod
    Commented Jun 19, 2014 at 14:11
  • 1
    @moose: Follow this chat...
    – Werner Mod
    Commented Jun 19, 2014 at 14:28
  • 1
    I've taken the liberty of editing your posting to make its purpose more immediately obvious. I've also changed some of the tags. Hope that's OK with you.
    – Mico
    Commented Jun 24, 2014 at 5:37
  • @Mico I went to your page, but I don't understand, what to do. I'm being asked to add an answer there, ah, what?
    – Keks Dose
    Commented Jun 24, 2014 at 8:00
  • @KeksDose - Umm, did you mean to address your query to moose?
    – Mico
    Commented Jun 24, 2014 at 8:39
  • @KeksDose: I need people who click on "Train" and then write the symbol. This can be done easily by clicking on a "Challenge". Does that answer your question?
    – Martin Thoma
    Commented Jun 24, 2014 at 15:16
  • @Mico: Thanks for editing my post!
    – Martin Thoma
    Commented Jun 24, 2014 at 15:18
  • @moose : when I tried to draw using an Ipad, the first stroke (even if I try to be very carefully) always seems to be randomly placed.
    – Sergio Parreiras
    Commented Jun 28, 2014 at 21:01
  • @SergioParreiras: Does it probably look like this: github.com/MartinThoma/write-math/issues/3?
    – Martin Thoma
    Commented Jun 28, 2014 at 21:19
  • @moose: yes, it looks exactly like the drawings in the link
    – Sergio Parreiras
    Commented Jun 28, 2014 at 21:41
  • Did you get data from Detexify after all? I just saw the ranking page has a new leader since I last looked at it.
    – Torbjørn T.
    Commented Sep 30, 2014 at 19:19

1 Answer 1

Reset to default
10

Have you thought of contacting Detexify? They should have or be able to collect the data you need. They might be willing to help.

2
  • I've contacted Daniel Kirsch (developer of Detexify) and got some very useful comments from him including his diploma thesis. But he could not send me the data.
    – Martin Thoma
    Commented Jun 25, 2014 at 15:06
  • @moose: Any plans to make your data public, so that others can also benefit from it
    – Aditya
    Commented Jun 30, 2014 at 12:35

You must log in to answer this question.

Not the answer you're looking for? Browse other questions tagged .