Which project to neural networks with text (linguistic or around) can be implemented with the student?

Colleagues, welcome.

The children are in school and one of them wants the project to get something "technical".
Learn to know and love (in Python e code a little know how) and I would like to offer them to dig in the direction of neural networks.
Material for literacy I'd saved, references to the literature to explore too.

Please advise where to start the training (what to take as the project). Preferably in working with texts (the second of the girls is fond of Philology).

And, in fact, any reference to adequate tutorials that will be able to understand the student of 10th class (high level math), I will be also grateful.
June 14th 19 at 20:22
5 answers
June 14th 19 at 20:24
You can try to do something like predicting scores for the composition based on age(class) of the author and the text itself works.

Well, to make it even more interesting, you can sharpen the neural network to predict scores for essays on the topics related only to the works of War and Peace, Tolstoy. For 10th class important.

Really need a place to get a large number of training data, works with various estimates, as dvoichnyi and those that perfectly.
June 14th 19 at 20:26
For a start, explain to your students what machine learning is and how it works. It is wise to start with the simplest linear regression and gradient descent, then move on to the task of classification and logistic regression, to explain why the linear model do not always(almost never). Next, tell us about an ordinary fully connected mesh and better optimization techniques(sgd, momentum, etc.)
After that, your students will be ready to meet with convolutional and recurrent networks. About word embeddings(w2d, bag-of-words, tf-idf etc.) you can tell right in the course.

Links:
1) https://github.com/goto-ru/Basic_ML, tasks designed for students in grades 10-11 and students of 1-2 courses.
2) word2vec
3) lab RNN: 1, 2.
4) karpathy.github.io/2015/05/21/rnn-effectiveness is a very clear about the recurrent network.
5) CNN in NLP: 1, 2, 3.
6) https://distill.pub/ - in simple language about complicated things.
Thanks for almost ready curriculum :) - aurore_Powlows commented on June 14th 19 at 20:29
It is certainly good to first rabotnici the basic Foundation of machine learning, but it's boring and tedious, I think it is better to start with a specific task for some specific topic, to interest children, and then to expand the horizons, if the theme they like. - Kaia.Bogisich96 commented on June 14th 19 at 20:32
that's your subjective opinion. I wasn't bored.

In any case, without understanding the basics of the type of optimization methods or the fact that the typewriter is a method of approximation, a good project to do. - lourdes_Bins50 commented on June 14th 19 at 20:35
heh, and here I am somehow the opposite.
Initially as a course project met the Kohonen self-organizing maps (SOM), and then got involved and their own rabbitanal classic fully connected multilayer perceptrons, then a few years later, when doing the parsing and classification, learned what word2vec, then the convolutional neural network and then machine learning, baesy, support vectors, recurrent network, generative, decision trees, boosting, ensembles etc. - Kaia.Bogisich96 commented on June 14th 19 at 20:38
June 14th 19 at 20:28
I have freshmen analyze the spam, the examples in the Internet a lot. Like, you know.

Still it is possible to upload jokes and compare them, for example, with random passages from literature and news. Download and random snippets cut they can try themselves. At the same time taste all the hardships of linguistic labor :)

In 11th class if they didn't run away, you can introduce them to word embeddings and neural networks feed not just key words, and words with meaning (of a vector, that is).
June 14th 19 at 20:30
Known to me, the easiest way to solve the recognition problem of the language of the document by comparing chastotnoi of characters (MSE) with known from the training corps.
June 14th 19 at 20:32
I would like to offer them to dig in the direction of neural networks.

any reference to adequate tutorials that will be able to understand the student 10th class

You myself - you HEAR me?!
PS: neural network is a formula lysmata.

UPD:
Topic: identification of all possible characteristics of word: part of speech, number, case, etc.
There's nothing particularly difficult if you don't go VERY DEEP. Need the basics of linala, statistics, and calculus. Specifically, to know what is the matrix vector product and be able to derived.
179 school, I suppose, is class 8 pass. - aurore_Powlows commented on June 14th 19 at 20:35
yeah, in Novosibirsk, 179 school we have up to 10 class was definitely and derivatives, and there are all sorts of bases of the towers... I think that education is enough to start to deal with neural networks, because after school, when I went to a CC, you already knew how to program the perceptron and back-propagation. - Kaia.Bogisich96 commented on June 14th 19 at 20:38

Find more questions by tags PythonNeural networks