Building AI for Good, By the People, For the People

Every second week we host a new challenge where a global community of AI engineers and enthusiasts collaborates to solve social problems through AI.

Rudradeb Mitra
Jun 23, 2019 · 8 min read

In the following, you can read about the true power of a community that goes beyond borders to solve impactful social problems through two-month AI challenges.

‘’A group of strangers from different corners of the Earth, who have never met each other before; transcending geographical borders and time zones to work together and solve fascinating social problems; whilst learning from and inspiring each other every single day! This isn’t just a figment of my imagination. Such a world exists and I am extremely grateful that I am part of such an extraordinary journey’ — Samir Sheriff, Omdena collaborator from India

The power of a collaborative environment

Imagine if the above-mentioned model works, how powerful that will be!

It will not only change education but also the future of work and how societies can come together to solve their challenges without relying on governments and large organizations. A true community-driven bottom-up approach.

Alone, we can do so little; together, we can do so much

— Helen Keller

Creating such an environment

Introducing Omdena’s Collaborative AI Challenges, where we host an impactful problem and a community of enthusiasts comes together to solve it. We are currently running four AI Challenges with over 600 AI engineers from over 66 countries.

In the following, we will share the results, after 4 weeks of work, from our first challenge.

The Challenge

Building a Machine Learning model for tree identification on satellite images for a Swedish startup. The solution will prevent power outages and fires sparked by falling trees and storms. This will save lives, reduce CO2 emissions, and improve infrastructure inspection.

The Results

Four weeks ago 35 AI enthusiasts who never met before, from 16 countries came together through our online platform. The community participants formed self-organized task groups and each task group either picked up a part or approach to solving the challenge.

Forming the Task Groups

The platform is a self-organized learning environment and after the first kick-off call, the enthusiasts started to co-operate. Below are screenshots of some of the discussions that took place in the first days of the project.

Soon different task groups were formed.

Task Group 1: Labeling

So far we got over 300 labeled images. A large group of people makes it not only faster but also more accurate through our peer-to-peer review process.

Active Participants: Leonardo Sanchez (Task Manager, Brazil), Arafat Bin Hossain (Bangladesh), Sim Keng Ying(Singapore), Alejandro Bautista Ramos (Mexico), Santosh kumar Pydipalli (India), Gerardo Duran (Mexico), Annie Tran (USA), Steven Parra Giraldo (Colombia), Bishwa Karki (Nepal), Isaac Rodríguez Bribiesca (Mexico).

Labeled images

Task Group 2: Generating images through GANs

Given a training set, GANs can be used to generate new data with the same features as the training set.

Active Participants: Santiago Hincapie-Potes (Task Manager, Colombia), Amit Singh (Task Manager for DCGAN, India), Ramon Ontiveros (Mexico), Steven Parra Giraldo (Colombia), Isaac Rodríguez (Mexico), Rafael Villca (Bolivia), Bishwa Karki (Nepal).

Output from GAN

Task Group 3: Generating elevation model

The task group is using a Digital Elevation Model and triangulated irregular network. Knowing the elevation of the land as well as trees will help us to assess risk potential tree posses to overhead cables.

Active Participants: Gabriel Garcia Ojeda (Mexico)

Task Group 4: Sharpening the images

A set of image processes has been built, different combinations of filters were used and a basic pipeline to automate the process was implemented to test out the combinations. All in order to preprocess the set of labeled images to achieve more accurate results with the AI models.

Active Participants: Lukasz Kaczmarek (Task Manager, Poland) Cristian Vargas (Mexico), Rodolfo Ferro (Mexico), Ramon Ontiveros (Mexico).

Output after sharpening

Task Group 5: Detecting trees through Masked R-CNN model

Mask R-CNN was built by the Facebook AI research team. The model generates a set of bounding boxes that possibly contain the trees. The second step is to color based on certainty.

Active Participants: Kathelyn Zelaya (Task Manager, USA), Annie Tran (USA), Shubhajit Das (India), Shafie Mukhre (USA).

Masked RCNN output

Task Group 6: Detecting trees through U-Net model

U-Net was initially used for biomedical image segmentation, but because of the good results it was able to achieve, U-Net is being applied in a variety of other tasks. It is one of the best network architecture for image segmentation. We applied the same architecture to identifying trees and got very encouraging results, even when trained with less than 50 images.

Active Participants: Pawel Pisarski (Task Manager, Canada), Arafat Bin Hossain (Bangladesh), Rodolfo Ferro (Mexico), Juan Manuel Ciro Torre (Colombia), Leonardo Sanchez (Brazil).

Output of tree detection through U-Net

What participants have to say

Within a week I learned so much about machine learning, satellite images, Google Earth Engine, data labeling and preprocessing’ — Amit Upreti, Omdena Challenge participant from Nepal.

I´ve already noticed that I learn new AI concepts faster within Omdena framework, but what I´m most surprised about is the speed of the progress that we are making on solving this difficult challenge. Only a couple of weeks in and I can already see some tangible results!’— Conrad, Omdena Challenge participant from England living in Mexico.

Now, how active is the group?

Active groups in the last 4 weeks. We had less than 20% dropout which is amazing!
Number of messages exchanged in 4 weeks

Additionally, we are building reusable tools for the World

We are not only solving a given problem but also developing useful, general, and extensible tools, as we care about sharing our work with the world. This is why we are working on packing up a set of utility functions that will help others to preprocess large image datasets before feeding an AI model.

And why not structuring this solution on a scalable automated process?

Rodolfo is working on `ImPipes`, a _pip-installable_ Python package to create image pipelines that will run a set of image processes, filters, and enhancements in a sequentially and very simple way (from one to a complete set of images!).

Why does this Collaborative model work?

The idea of collaboration is nothing new. Classes in business school were taught on the future of work, crowdsourcing knowledge work, bounties/x-prizes, new organizational structures, etc. There were a lot of proofs, but it was never really successfully implemented. Now is the time to do this due to the following reasons:

  • Tapping into a large pool of people with high intrinsic motivation: Online education has made it possible for anyone, anywhere to gain knowledge. What we have seen over and over again that people applying and participating in these challenges have high intrinsic motivation. As mentioned by Chidinma Obiagwu, Research Assistant at Hitachi — ‘I always wanted to make an impact, and be relevant. I believe one can’t measure the good of education if you can’t use it to solve social problems’.
  • Eagerness to collaborate and not to compete: We have been always told to compete with each other to be on top. But we hardly realize the power of collaboration. What we were able to achieve in 4 weeks is truly amazing. No one really knows what model will work best so dividing the problem and trying different approaches makes a lot of sense in the world of AI. Different groups can tackle the problem in different ways while still collaborating and making the whole process faster.
Different groups trying different models
Quality control is done through a peer-to-peer review process
  • Accessible tools, streamlined processes for shared learning, mentorship, and incentive structures: We use the best-in-class available tools and processes. Our processes are made for shared learning where everyone benefits. Kanwal Shariq from Pakistan asked to remain an observer in the group, ‘I am learning a lot just by seeing how they are collaborating and working the problem and it is really eye-opening’. We also have incentive structures and mentorship in place to create the added value. We use concepts from the tournament theory, and thus do not have compensation based on individual performance because it fosters undue competition among contestants.
Members presenting their work to the community

From outside, it all may seem chaos but as pointed out by one of the participants

‘What seems chaos at first, one starts to appreciate the experience and realize the relevance of working in self-organizing environment, where crucial conclusions and optimal solutions eventually emerge as winners. It is very well welcomed and valuable experience. It makes me proud to be a part of it.’- Vjeko Hofman, Omdena Challenge participant from Croatia living in Poland.

Our challenges at Omdena can be tough and learning AI and Data Science is demanding. But it is so much fun to do with people from all over the world while solving a social problem.

One of Omdena’s participants, Łukasz Murawski says, Omdena gives an opportunity to contribute, to do some good that matters and lasts… far beyond our lifetime. Corporations will not do it. I believe regular, passionate people like You and me will!’

Host your AI challenge with us

If you have an idea that uses AI and will benefit society, contact us here. If selected, we will get a community of AI experts and enthusiasts behind it and solve the problem.

If you want to receive updates on our AI Challenges, get expert interviews, and practical tips to boost your AI skills, subscribe to our newsletter.

We are also on Facebook, LinkedIn, and Twitter.


Building Real-World AI Solutions Collaboratively