Beirut AI Community Recap: August 2019 (Bootcamp Edition)
The long-awaited August came and went. It was truly one for the books. The Beirut AI Bootcamp finally made its way into the light and it was beyond our wildest dreams.
As usual, before anything else, we want to thank you for all your support. Our goal has always been to make AI accessible and applicable to anyone and everyone. Without your support, we could never say we have a community of people who want to improve and help others do the same in the world of AI.
And, without further ado…
Let’s Walk You Through Our Adventure (a.k.a. The Bootcamp!)
We know, from experience, how easy it is to forget a concept, especially in AI, when you don’t put it into practice. No matter how cool or fascinating an idea is, if you don’t try it out yourself, it is hard to retain it. Building upon that concept we designed our five-day bootcamp to include as many hands-on projects as we can fit into it. From day one we made sure every student left the venue having practiced the theory that was provided. We also wanted every participant to leave this experience with a feeling of pride and accomplishment. That’s why on the last day we organized a hack day competition where each team, within the span of fewer than 24 hours, was expected to put to the test the skills they acquired to devise projects that pique their interest.
What a road to discovery it was. From trying to solve the waste problem to creating an AI that generated its own horror movie scripts, it was an entire galaxy of creativity, discovery and innovation. Our participants, known more formally as astronauts, explored the many planets of the AI galaxy.
Our six days of AI galaxy exploration ranged from introduction to innovation and this is how they went:
Planetary Visit I: Data Science
We wanted to make sure all our participants were armed with the right tools to conquer all the days to come. We started off by introducing Python for Data Science. Every participant was given a chance to get accustomed to the basic tools, libraries and modules they would need to conquer the AI journey.
The participants’ comfort with the tools was put to the test that afternoon. The first hands-on project of the bootcamp exposed our participants to a real-estate problem. As with any AI project we start off with understanding the business problem and need. Once that is complete we can head into actually exploring, visualizing, cleaning and preparing a dataset that we obtain. Once we’ve completed the previous steps as well as extracted useful descriptive features, we run our predictive models. The models were compared, evaluated and improved as we understood how to refine their performance.
In this first planetary exploration, we built an intuition around how data science projects flow. We ensured that every participant had a clear grasp on the foundational tools as well as the basic techniques and methods used to solve any AI problem.
Planetary Visit II: Deep Learning
The second day of the bootcamp started off with an introduction to the famous Neural Networks. We discussed how they are built and how they learn from data. We promptly moved on to practice with an introduction to Keras with Tensorflow. Participants learned how to build their first neural networks with Keras, how to pass the data to the networks, and how to train and evaluate the networks’ performance. We explored different use cases of regression, binary classification, and multi-class classification and investigated how the choice of application affects the choice of the loss function and output layer designs.
The second half of the day proceeded with more advanced concepts such as K-Fold cross-validation of data, building Keras pipelines, configuring Keras callbacks for model checkpoints, early stopping and dynamically adjusting the learning rate.
It was a challenging day with a vast amount of information to explore and process, but our participants survived with great enthusiasm and grit!
Planetary Visit III: Computer Vision
For the third planetary visit, the students were prepped and ready to explore a whole new world, Computer Vision. We defined the idea of an image to a computer, which seems so trivial to humans. We explored how this data type can be manipulated with a vast array for linear algebraic, statistical, and calculus driven algorithms. We looked into edge detection, color gradients, Gaussian smoothing, rotation matrices, and more! We spent the majority of the morning working on computer vision-based mathematics, from matrix transformations to (and especially) convolutional transformations. Participants were introduced to OpenCV and how it is commonly used in practice to assist in implementing complex methods with just a few lines of code.
After completing the OpenCV linear and statistical techniques section, for the afternoon we deepened the astronauts’ understanding of deep learning for Computer Vision. After creating the basis at the start of the day, and building up from the first two days, we discussed the nature of convolutional neural networks (CNNs). We closely examined several pre-built models that won successive ImageNet competitions from 2012 through 2016. With this we clarified the evolution on neural networks. This unique path through the successful history of CNNs provided an understanding of why we design CNNs the way we do today; but more importantly, how we can capitalize on the open-source communities’ pre-built CNNs using ImageNet.
The material of Day 3 was expansive and detailed. But the students were enthusiastic, continuously probing the instructor with very interesting, targeted, and important questions.
Planetary Visit IV: Natural Language Processing
On day 4, we traveled to a planet far far away; Natural Language Processing. Unlike other AI planets, NLP is recently discovered, and there is still a lot to unravel. Our first station was setting the ground with basic definitions (what is NLP?) and applications (why it is significant). We looked at why NLP is a hard problem and what the main NLP challenges are. We discovered one of the most mesmerizing works of art in NLP, feature extraction (aka. text vectorization) specifically Bag of Words & Word Embeddings.
As always, theory is great but nothing beats putting it into practice. We applied the NLP knowledge we learnt during the journey into 3 different applications: (1) sentiment analysis to extract the writers’ emotions from digital text, (2) auto-completion feature in chats, and (3) text translation from English to French.
We left this planet full of knowledge and enthusiasm! As an ending note, a personal visit to a top NLP conference, ACL, was shared with the participants where they heard about the latest NLP trends & state-of-the-art models by NLP giants such as Google, Facebook, Amazon, etc.
Planetary Visit V: Time-series Analysis / Recommender Systems
Our fifth exploration involved a surprise; two AI topics in one day! The day kicked off with an introduction and understanding of one of the most commonly found data types, time-series. We explored the nature of these sequences and looked at the process of collecting, reading, visualizing, and understanding the underlying behavior of time-series data. Our goal for this planetary visit was to ensure that our participants can comfortably work with time-series data as build functional and accurate predictive models. We passed on this information by providing a hands-on experience. Participants created a project where they were predicting air pollution levels given daily pollution levels measured in the city from previous years.
As a twist on our final day, we covered our last topic for the bootcamp: recommender systems!
Very few people haven’t heard of Netflix! And those who have heard of it know exactly how on the nose Netflix’s movie and show recommendations can be! But that’s not the only example, from shopping to music companies recommend us products we would love to buy. From those real world examples we introduced the students to where and how the most common recommender systems are used. We did a walk-through exploration of a full movie recommendation implementation. Finally, the participants got the chance to implement a new recommender system engine from scratch!
By the end of the last day, our astronauts were familiar with the foundations of Data Science and its tools. They became comfortable with dealing with various data types from tabular data, images, text, to sensor (time-series) data. Our AI heroes were armed with theoretical discoveries as well as solid experimental practice, and they were ready to be released to examine a solution to one last problem of their choice on the last day of the bootcamp.
Galaxy Exit VI: HACK DAY
The final day of the bootcamp put our AI heroes to the test! Participants were asked to form teams of 5 and build their own AI solution in only 6 hours. It was definitely a challenge, but we were proud to see our students succeeded in their challenge and present amazing ideas and results at the final pitches.
Topics ranged widely and included ideas of equity price prediction, low-cost garbage sorters, cardiovascular disease prediction, detection of diabetic retinopathy, car model classifications, detecting SKU’s from store shelves, learning personalized user experiences, generating horror storylines, image captioning, steel defect detection, and handwriting translators.
An amazing amount of effort and knowledge was shown by all our participants. To make the experience more exciting, we announced that the top 3 teams will be taking home some prizes!
We believe everyone did an incredible job! And we can’t be happier with the effort and display of knowledge. Everyone was a winner to us, but the teams who won the prizes are:
First Prize Winners: Jaydoo
Second Prize Winners: Dash
Third Prize Winners: Our work is garbage
A Message To Our Sponsors, Partners and Judges
We couldn’t have done it without you and your incredible support!
From our team to yours, Paragon Shift, Zaka, and Mindsets… Thank you! Without your generous support we would not have been able to provide the experience we envisioned providing.
We also cannot thank our partners enough for providing everyone with a comfortable and enjoyable experience throughout the bootcamp from facilitating transportation to food and coffee accommodations.
To the judges, Ali Tarraf from Mindsets, Aly Harakeh from Microsoft, and Raymond Younan from Paragon Shift. You took time out of your Saturday to assess the projects generated from the hard work of our participants. You gave us feedback and support that we truly appreciate! Thank you!
A special thanks to Microsoft for providing our participants with certificates of completion and to the American University of Beirut’s Electrical and Computer Engineering department for providing the bootcamp venue and technical support.
Finally, thank you to the incredible Beirut AI team for making this bootcamp truly memorable. To the volunteers, instructors and organizing team, an immense thank you!
“Alone we can do so little; together we can do so much.” — Helen Keller
We Want to Make it Even Better
Our participants weren’t the only ones learning in those five days. We were too. We want the bootcamp experience to only get better from here. Our goal for the bootcamp is to offer the ultimate experience of maximum knowledge gain and expertise with minimal time and cost. That’s why it’s important for us to hear from you.
We would love to know what you think will make for an incredible bootcamp. Just click here to share your thoughts with us!
Wanna go on an AI Adventure?
This will definitely not be our last bootcamp, and we want as many of you to get the chance to experience it as possible. If you’re interested in exploring the vast galaxy of AI through gaining hands-on experience, we want to see you there.
Our next AI mission is under construction! If you want to be the first to know when the next Beirut AI Bootcamp begins, sign up here.
We can’t wait to see you!
Like what we’re doing? Give us a clap!
Any topics you would like us to cover?
Drop a comment or send us an email at beirut@city.ai
You can even join us on slack
For more information visit city.ai/cities/beirut.
