AI Projects #6 for Social Good

AI Projects #6 for Social Good came to an end on May 8 with a public showcase that welcomed 108 AI enthusiasts. We were happy to meet with members of our AI community, both from academia and industry, who joined us to witness the results of  5 teams’ 3-month of hard work.

Elaborating on the ways to create societal impact, our 21 participants teamed up to develop a variety of projects focusing on topics such as medical applications, Turkish NLP, and demographic analysis using object detection. While we are happy that 4 out of 5 teams decided to turn their projects into academic papers to be submitted to various AI conferences in line with their areas of research, we are happier to witness how the idea of social good helped our AI community think outside of the box and tackle more unusual challenges, different than anything else that was discussed in our previous batches. 

Another great news about this batch is that we launched a demo website, where everyone can play with some of the projects that were held under inzva AI Projects Program.

Moreover, our showcase videos are already up on our YouTube channel, and teams’ presentations can be found here.

AI Projects #6 PROJECTS

1- Turkish Offensive Language Detection

Uras Mutlu, Ekrem Bal, Talha Çolakoğlu, Abdurrahman Dilmac

Modern NLP tools with deep learning can be used to build a strong model that detects offensive language. The team aimed to reproduce the OffensEval 2020 benchmark on the Turkish offensive language detection task and improve the model with different techniques such as text normalization and data augmentation. They plan to finalize the project as an open-source API that interested people can contribute and anyone can use in their applications or websites. 

The team performed a demo of their work in the showcase, and uploaded it to our demo website for other AI enthusiasts to try.

2- Optimal Data Augmentation for Biomedical Image Segmentation

Billur Selin Zaza, Enes Sadi Uysal, Mehmet Yiğit Özgenç, Şafak Bilici, Onur Boyar

This study aimed to find the optimal data augmentation volume and optimal data augmentation strategies in the Retinal Vessel Segmentation problem. 

Data augmentation plays a crucial role in performance of most of the image classification on the image segmentation tasks. When there is a shortage in the training samples, many applications rely on data augmentation techniques to increase the performance of the model. The success of the Retinal Vessel Segmentation is crucial to diagnose various diseases like diabetes, cardiovascular diseases, etc. This team proposes a data augmentation strategy for this problem which outperforms the previous studies using the same model architecture.

This project was chosen as our participants’ favorite in our showcase and a paper based on it was already uploaded to arXiv. 

Codes can be found on our GitHub page.

3- Building Knowledge Graph based on Turkish Wikipedia

Alaeddin Selçuk Gürel, Ateş Bilgin, Mustafa Barış Çamlı, Okan Çiftçi

Knowledge graphs are basically a way to store knowledge on a graph form, using entities, relationships and attributes. The attributes and relations between entities are stored in graph format. Knowledge graphs are useful in information retrieval, QA, or Decision Support tasks. In that project, the team constructed a knowledge graph using a not-so-common language data source which is Turkish Wikipedia and evaluated it on Graph tasks. Finally, they constructed different graph embeddings, and compared their results. A highly academic work greatly contributing to the Turkish NLP literature, this team believes it will benefit the Turkish AI community in the long run. 

Watch their showcase video, then check out the codes to learn more about the project.

4- Medical diagnosis decision support system for Parkinson's disease

Can Bulguoğlu, Duygu Ay, Handenur Çalışkan, Orhan Ağaoğlu

For generations, assessment of speech abnormalities in neurodegenerative disorders such as Parkinson's disease (PD) has been limited to perceptual testing or user-controlled laboratory analysis based on very small human vocalization samples. However, using speech signals, detection of Parkinson's disease is now possible with the help of machine learning and deep learning methods. In this project, the team aimed to detect Parkinson's disease by processing speech signals and reach the state-of-the-art level. Furthermore, they wanted to investigate whether our model can achieve a generalized performance for the voice signals from different languages.

While this team will submit a conference paper based on this project, they have already made a demo of their work. 

To deepen your understanding, you can check their showcase video.

5- Object Detection with Street View Images for Demographic Analysis

Alara Hergün, Başak Ekinci, Efehan Danışman, Sefa Kurtipek

Previous studies made in other countries proved that a huge amount of resources are allocated for data collection to estimate the demographic structures of the neighborhoods. Similarly, this team used object detection techniques using Google Street View Images of Istanbul to predict the socio-economic development levels of each neighborhood. They determined objects in street view images such as cars, trees, and miscellaneous street items to predict the socio-economic development level of the neighborhood. Mahallem Istanbul project’s results are selected as the target value to predict by using street view images. This team decided to continue their project within the scope of our next batch!

Do not forget to see their codes, go over their showcase video.

Moreover, the team wrote a great blog post about their journey, check out here.

As we came to the end of our sixth showcase we are excited to take AI Projects further by focusing on different areas of interest. It can be health, industry, earth,, and many other things. 

We thank all our teams for their effort and wish to welcome other like-minded AI enthusiasts to our next batch, which will be held in October.

inzva is supported by BEV Foundation, an education foundation for the digital native generation which aims to build communities that foster peer-learning and encourage mastery through one-to-one mentorship.

Subscribe to our newsletter here and stay tuned for more hacker-driven activities.