Critical thinking is a 21st Century Skill and requires in-depth analysis of a course topic that allows students to see different sides of the issue, which they have to defend. I have done some work on debating with much success, so I want to suggest that you can look at this area as students appear to lacking in this skill:

https://www.researchgate.net/publication/274720155_Debating_A_Dynamic_Teaching_Strategy_for_Motivating_Students_and_Teachers

Many thanks,

Debra

Conference Paper Debating: A Dynamic Teaching Strategy for Motivating Student...

thank you for your suggestions

Görüntü işleme ile ilgili çalışabilirsin, özellikle jenerik obje tanıma ya da insan yüzü açık alanlar.

Görüntü İşleme alanında çalışmak istiyorum zaten tunç hocam. Danışmanım Image caption olabilir demişti sanırım jenerik obje tanıma ile aynı alan oluyor.   Yada Göz den elde edilen görüntülerle şeker hastalığının hangi seviyede olduguna bakabilirsin demişti.  Cevabınız için teşekkür ederim Tunc hocam.Thank you harshvardhan for your reply

The trend now is to link deep learning and artificial intelligence to neuroscience. The goal is to develop neuroscience- inspired artificial intelligence models. follow these links for more information:

http://www.cell.com/current-biology/fulltext/S0960-9822(14)01039-2

http://www.sciencedirect.com/science/article/pii/S1077314216300339

http://ieeexplore.ieee.org/xpl/login.jsp?tp=&arnumber=7458013&url=http%3A%2F%2Fieeexplore.ieee.org%2Fxpls%2Fabs_all.jsp%3Farnumber%3D7458013

A severe limitation of deep learning is the lack of explainability. These methods are black boxes and even the designers cannot explain why do they work. Therefore a good subject would be adding explanability modules to deep learning systems.

Is there any sample code (python or java) for using and experimenting deep learning ?

There are many deep learning frameworks that are easy to use and do not require high programming skills. However, some of them requires a GPU installed in your computer. Deep learning frameworks such as:

-Cuda- convnets: https://github.com/akrizhevsky/cuda-convnet2

The back end is Cuda/C++ and the front end is Python

-Pylearn: developed in Python

-Torch7: developed and open sourced by Facebook

- TensorFlow: developed by google

- Theano

- Cafe

@Mohamed Zarka,  Do you play board game 'go/weiqi/baduk' ?  With that game experience and with this thread about 'convexity cost' etc... I think the two should give you better ideas at experimenting deep learning thought process, aside from learning a modern computing languages.  That is my two cents. 

https://www.researchgate.net/post/What_are_applications_of_convex_sets_and_the_notion_of_convexity_in_mathematics_and_science

https://www.researchgate.net/post/Do_you_see_Googles_DeepMinds_AlphaGo_defeat_of_professional_9-dan_Go_player_as_good_as_IT_gets

thank you very much for your suggestions. this topics are very interesting especiially image caption.

Hello Sezer ali Aksu,

I thought  these links may be useful to find current .

http://deeplearning.net/

https://www.cs.toronto.edu/~gdahl/papers/Dahl_George_E_201506_PhD_thesis.pdf

http://nlp.stanford.edu/~socherr/thesis.pdf

http://upcommons.upc.edu/bitstream/handle/2099.1/20268/main.pdf

http://lib.tkk.fi/Diss/2014/isbn9789526055756/isbn9789526055756.pdf

https://www.cir.meduniwien.ac.at/assets/Uploads/Masterthesis-SeeboeckPhilipp-Version28-03-2015.pdf

http://ir.uiowa.edu/cgi/viewcontent.cgi?article=3511&context=etd

http://scholarworks.rit.edu/cgi/viewcontent.cgi?article=9984&context=theses

Select your area of investigation and then go from the beging and basic, master on it and then drive in deep knowledge.

thank your abdul and subrata. İmage caption  interesting and ı want to study about image caption with deep learning subrata. Is it interesting?

It is an very eficent learnig algorithm that return a important score for classification and recognitiontasks. You mast chose an area for investiging de deep and you must have a large dataset to doing your experementation.

my topic ,which ı choose, image caption

Good choice, for it is still a challenge.I t is a mixture between image recognition and natural language processing. Less accuracy has been achieved  in image captioning compared to either field alone.

thank you for your reply marwa

You are welcome Sezer

Hi Sezer and everyone,

I've said "image processing" in my earlier comment but it's a huge area with lots of different sub domains.

You can work on multispectral images for blood analysis, remote sensing, chemical analysis etc. In these problems, the data is coming from a wide range of wavelengths and sometimes it's not easy to find the proper calculations by examining these wavelengths one by one. So, one can use neural networks to reveal the structures in the data. And these areas have high importance yet they are not studied as much as face recognition or object classification etc.

Recurrent network structure is another possibility. In so many domains, the data is time dependent which means you have to consider the data sequences for coming up with a solution. For example, the amount of electricity used in a city on a specific date is an important question and you can apply recurrent solutions to these areas as well. Market size predictions, resource related calculations, population modellings, migration modellings etc. these are all relatively open areas for neural network solutions.

One can also study the shortcomings of neural networks. Are they really learning? If they do so, is it really deep? These questions have scientific aspects as well as philosophical ones. Why, for example, we are able to identify/recognize human faces with more than 99.9% success but we still have no idea how these systems can be generalized to other domains? You might know that the "generalized AI" approach is ascending again but this time with the help of statistical methods. Maybe instead of working on some specific questions, you might want to be a part of the research community that is trying to connect all these methods together to generate some organic-like intelligence.

One might also consider the philosophical questions. What is that thing those neural networks are doing? It is not deduction for sure. It has some inductive shades as well, that's also for sure. But what else? Do they perform some form of abduction? Can they create new lexicons or concepts? Can they abstract over different problem domains? Can they go beyond statistics? Or humans, can we go beyond statistics?

Have a good day everyone.