AMLD Africa 2021

Conference Opening

[Invited Keynote] Reimagining a More Inclusive Future with AI

[Invited Keynote] - The Great Potential of AI to Address Global Health Challenges

The COVID-19 pandemic has elucidated a number of challenges and the need for adaptive health systems worldwide, particularly when there is a “shock,” such as an outbreak or pandemic, to the health system. This talk will cover examples of AI used to address Global Health and focus on recently published work on COVID-19 in Nature - Scientific data and on our Worldwide Non-pharmaceutical Interventions Tracker for COVID-19 (WNTRAC).

Tunisian AI Tech to fight covid19 with limited resource

The talk will be about :
Using AI to for fast and mass detection of covid19 GGL using underutilized medical infrastructure in Tunisian such as CT Scan machines.
Lung cancer diagnosis and monitoring by hybrid augmented intelligence.

Automatic image analysis and artificial intelligence for cervical cancer detection with a smartphone-based solution

Cervical cancer is the fourth most common cancer in women worldwide. Both the incidence and the mortality of this preventable and treatable disease are higher in low- and medium-income countries. This situation could be avoided with a wider access to early detection tools. We propose a smartphone-based solution that automatically detects cervical precancer and cancer from videos of the cervix collected during a routine screening method called Visual Inspection with Acetic Acid (VIA). The introduction of computer-aided diagnosis such as our automated VIA classifier test may play an essential role as a rapid, affordable and efficient tool for large-scale screening.

[Panel Discussion] - How AI is helping health reach the unreachable

With Emilian Popa, Eduard Booty & Valeria De Luca.
A panel discussion between two startups levering AI and delivering healthcare solutions in Africa. Ilara Health makes common diagnostics accessible and affordable to the 500 million people in Africa who struggle to access even basic blood tests today. Reach52's health tech platform enables access to a full range of health services in low-connectivity markets - from equipping health workers, screening and medical advice to a marketplace of affordable products, last-mile logistics and data analytics

[Invited Keynote] - Measuring and optimizing crop production using aerial imagery and machine learning

Aerobotics is helping farmers around the world optimize their crop performance through analytics derived from aerial (drone & satellite) imagery. This talk will dive into the technology and learnings that Aerobotics has developed within the industry.

Predicting Regional Locust Swarm Distribution with Recurrent Neural Networks

Locust invasion issue is considered as one of food insecurity reasons around the world including Africa, Asia and Middle East since antiquity. Huge numbers of desert locusts are swarming across the affected regions and increase under breeding conditions. It can affect the health and the lives of millions of people. Different ways have been used to reduce the effect of locust invasion, chemicals to prevent swarm formation, satellites, and sensors to detect and monitor locust breeding areas. But these methods have limitations since they have not been able to put down the upgrowth and the mass behavior of locusts. In this study, we utilize machine learning to predict the location and density of locust swarms in advance using the available data published by the Food and Agriculture Organization of the United Nations and published on the Locust hub portal. The data consist of the location of the observed swarms and environmental information, including soil moisture and the density of vegetation. The results show that our model can predict the location of locust swarms and the expected level of damage using the density notion successfully.

Developing localized data models for smallholder farmers

Smallholder farmers produce 1/3 of the world’s food, and up to 80% of food in regions like Sub Saharan Africa and Asia. Lack of data and fragmented markets creates challenges of seamlessly integrating smallholder farmers to a digital world. A scaleable model that captures smallholder farmers’ data is critical in developing AI and ML models that ensure food security.

Developing and deploying data-efficient land classification models for agriculture

With Ivan Zvonkov & Gabriel Tseng.
NASA Harvest supports food security around the world by using satellite observations to improve our understanding of agriculture. This talk will showcase the models Harvest has developed for low-data regions, and the cloud infrastructure we have built around these models to ensure their scalable deployment.

Tech- No Superhero Machine Learning, Africa and Sustainable Ag.

Sub Saharan agriculture is a critical sector to the continent and internationally. It is also broken in fundamental ways. Machine learning can address some of these issues, however a tech first approach will not work and the continent is littered with failed attempts of those who tried.

A simple NLP Chatbot for obtaining real-time farm production data at scale via SMS

With smartphone penetration across rural Sub-Saharan Africa still lagging behind that of simple phones, low internet connectivity remains a limiting factor to the exponential growth of online platforms. Farmster has developed a simple chatbot to enable rural farmers to join a free, online classified for farm produce without needing to use mobile data or access a smartphone. I share how our chatbot is simple enough for user experience and robust enough to gather information from farmers, and outline remaining challenges for human-machine interactions among least-reached demographics.

[Invited Keynote] - Earth Observations and Machine Learning for Food Security in Africa

Model Predictive Control for Microgrid applications in Africa

Microgrids are an integrated energy system consisting of a group interconnected Distributed Energy Resources (DERs) within clearly defined electrical boundaries that act as a single controllable entity with respect to the grid. To operate such microgrids, state-of-the-art energy management systems mostly rely on model predictive control (MPC), where the site's load and photovoltaic production are forecasted over a prediction horizon of typically 24-48 hours based on which an optimal control plan is computed for the flexible DERs. Schneider Electric's microgrid energy management solutions enable to optimize the operation of the on-site energy resources and loads using MPC algorithms. Microgrids are gaining popularity in Africa as an alternative to conventional power grids and a cost-effective way to enhance the access to energy.

Mapping of Africa’s Built Environment using remote sensing

Presentation of the Google Research team in Ghana work that uses high-resolution satellite imagery and computer vision to calculate continent-wide data for Africa. The presentation will go over labeling and modeling techniques used, the footprint, and potential social good use for the footprint generated.

AI for Low-Cost Urban Air Pollution Monitoring

The presentation will start with an introduction to Air Pollution in North Africa, followed by the MOREAIR project and it's strategy based on medical data, IoT and AI for monitoring and predicting air quality in urban areas.

The next generation of African infrastructures, challenges & opportunities: a computer vision case study

DeepFlows is developing a cloud-based traffic intelligence platform to help transport planners, governments and city councils reduce the friction to access traffic data using computer vision technology. This presentation will walk the audience through the opportunities that we are visioning and the challenges that we are encountering when it comes to applying machine learning to collect traffic data in Tunisia. DeepFlows is a Tunisian based DeepTech startup with the ambition of becoming the next African unicorn in advanced traffic management systems.

Will Africa leapfrog to renewable energy this decade?

Energy scenarios, relying on wide-ranging assumptions about the future, do not always adequately reflect the lock-in risks caused by planned power-generation projects and the uncertainty around their chances of realisation. In a study recently published in Nature Energy, we built a machine-learning model that predicts power-generation project failure and success using the largest dataset on historic and planned power plants available for Africa, combined with country-level characteristics. We found that the most relevant factors for successful commissioning of past projects are at plant level: capacity, fuel, ownership and connection type. We applied the trained model to predict the realisation of the current project pipeline. Contrary to rapid transition scenarios, our results show that the share of non-hydro renewables in electricity generation is likely to remain below 10% in 2030, despite total generation more than doubling. These findings point to high carbon lock-in risks for Africa, unless a rapid decarbonization shock occurs leading to large-scale cancellation of the fossil fuel plants currently in the pipeline.

Leveraging machine learning to achieve cleaner air in African cities.

In this talk, I will share the experience from AirQo (https://airqo.africa) on the development of low-cost sensing technology and applying machine learning to advance air quality monitoring, modelling and analysis in African cities. I will also talk about how we are leveraging digital technologies to achieve and inform interventions that contribute to clean air actions.

[Invited Keynote] - Measuring Economic Development from Space with Machine Learning

[Panel Discussion] - How AI is changing the world of low carbon electrical energy systems.

With Shinez Chalabi, Rosina Norton & Elizabeth Nyeko.
A panel discussion between two startups levering AI in the area of renewables and cleantech in Africa. M-KOPA is a connected asset financing platform that offers millions of underbanked customers access to life-enhancing products and services. Modularity Grid is decarbonizing and strengthening energy systems, with the end goal of accelerating climate-conscious development.

[Invited Keynote] - A Future of Shared AI Knowledge

Strategies to apply speech recognition in low resource language : WOLOF use case

In this talk we will discuss the difficulties and solutions of using Machine Learning techniques to develop ASR models in the context of low resource languages. We will talk about the different strategies of collecting data and leveraging non labeled data and finally show a demo with WOLOF use case.

Tunisian dialect NLP journey as low-resourced language

While being so diverse and rich, Arabic language and particularly Arabic dialects are still under represented and not yet fully exploited by deep learning because of the lack of available data. In this presentation, we introduce a methodology of collecting, cleaning and preprocessing Tunisian dialect data in order to create a language model and fine-tuning it for different NLP tasks

[Invited Keynote] AI Language Models - Latest Developments

[Panel Discussion] - About NLP and the African context featuring Instadeep and Rural Sense

With Lauren Newbould, Nourchene Ferchichi & Abdoulaye Diack.
A panel discussion between two startups leveraging NLP to deliver value in an African context. Rural Senses is using AI-enhanced data analysis and user-centric data collection, to make it easy for project developers to understand rural community needs and evaluate project impact, efficiently and accurately. Instadeep made understanding the Tunisian dialect by machines now possible with TunBERT a joint collaboration with iCompass.

[Invited Keynote] - Setting up a Research Cloud for Africa

I will describe the research cloud infrastructure set up by the Inter-University Institute for Data Intensive Astronomy (IDIA), a partnership of three South African universities and the South African Radio Astronomy Observatory. The institute is designed to enable African researchers to take the lead in large survey projects and other observations from the MeerKAT Radio Telescope. The MeerKAT is a precursor telescope to the international Square Kilometre Array radio telescope and will be integrated into its phase I deployment. The scale of the data coming from the MeerKAT requires a paradigm shift in astronomy research, led by enabling institutions like IDIA and by progress in machine learning applications. I will present some of our work on the data processing pipeline, on data visualisation and other research, as well as our skills and capacity development programme. The research cloud facility itself also serves the bioinformatics community, with African genomics at the forefront.

Enhancing African Machine Learning at a low cost

Despite the race towards Artificial Intelligence Research, Africa provides a limited scholarly output on Machine Learning from quantitative and qualitative perspectives. In this research presentation, I introduce several hints and methods that can be applied by local African scientists to simplify, enhance and validate their Machine Learning applications at a limited cost. The approaches that will be discussed mainly rely on open resources, simple algorithms and semantic web applications to achieve their goals.

Zindi and the power of ML for good

From a good idea, Zindi has grown into an engine for putting untapped data science talent to work across borders and across sectors. The 25 000-strong Zindi data science community has, over the last 2 years, produced remarkable machine learning solutions to real African problems, proving the power of African talent to change the world. In this presentation, we will talk about how Zindi-curated models have had an impact in fields as diverse as logistics, agriculture, health, climate change, and finance, and how the platform is building and growing into a world class community of practice in data science.

The Deep Learning Indaba Mentorship Programme

The Deep Learning Indaba Mentorship Programme aims to strengthen the African Machine Learning (ML) community via the development of fundamental skills. The Programme matches community members with mentors for short-term personalised interactions across a wide range of topic areas. Website: https://deeplearningindaba.com/mentorship/

GESDA Roundtable: Anticipating the future of ML and its impact

With Nanjira Sambuli, Rüdiger Urbanke, Martin Müller, Muthoni Wanyoike, Fabrice Lauper & Alexandre Fasel. Anticipating the future of ML and its impact The Geneva Science and Diplomacy Anticipator (GESDA) is a young foundation initiated by the Swiss Federal Government. Its mission is to engage in science diplomacy and strengthen the international Geneva ecosystem by anticipating future scientific and technological advances and their impact on people, society and the planet. GESDA’s ambition is to put early on cutting-edge science at the core of the multilateral agenda and support the development and deployment of global initiatives for the common good. During the past year, GESDA has convened scientific leaders in the field of machine learning and artificial intelligence to anticipate developments in the field with a 25-year perspectives. One of those conversations let to a dedicated track at the AMLD 2021 series on 10 May where we identified the major science and engineering trends, the main bottlenecks and had a first discussion around the implications of those developments on society and global governance. This roundtable will expand previous debates and current discussions between science and diplomacy leaders and bring in the perspectives from Africa. The main questions to address: What are the major developments in machine learning and artificial intelligence that will have an impact in Africa? What are the trends and which approaches make a difference? How to overcome the divide between the actors who currently design the technology and the people who will use it or be affected by it? How can science diplomacy contribute to reduce the divide? What governance challenges exist around machine learning and artificial intelligence to ensure a development, deployment and use of the technology that benefits African societies?

[Invited Keynote] - AI, The past and a Future

Machine learning for preserving heritage with Google Arts & Culture

Advancements in computer vision, machine learning and cloud technology have enabled new opportunities for the cultural sector in exciting ways. With partners from around the world, Google Arts & Culture has launched new initiatives to help promote, share and preserve heritage, from the ancient world to present-day cultures. Fabricius is a new online tool using machine learning to help everyone write Egyptian hieroglyphs and for researchers in particular during the translation process. This educational tool is accessible in Arabic and English and also provides a step-by-step tutorial to learning more about Egyptian hieroglyphs. Woolaroo is an experimental web app using the Cloud Vision API to help preserve and promote endangered languages from around the world. Users can simply take a picture of the objects around them and learn the words -- and listen to them pronounced -- in 10 languages. Woolaroo is also accessible in French, Spanish, Arabic, Italian and Hindi.

The current state of AI in Tunisia

Machine Learning as a field has seen an explosion of interest in Tunisia in the past 2-3 years, with the number of companies, communities, content creators, and training sessions skyrocketing. For an external observer, it might not be easy to gauge the overall maturity of this field in Tunisia, which is why an overview of its current level of adoption in the different aspects of the country is needed. That is why we will explore the current Tunisian ML landscape, from Government to companies, passing by communities, research, content, market... and showcasing the biggest breakthroughs and latest statistics.

[Invited Keynote] - Taking AI from the Lab to the Real World

We will discuss the challenges and opportunities with actually applying AI and ML to add value to businesses, with a focus on Africa.

AI for business impact : the curse of Kaggle

The success of Kaggle those last years has enabled hundreds of thousands of Data Scientists to improve their skills in Machine Learning, thanks to their famous competitions. But the problem is that it has also led a lot of people to think that "building the best algorithm" is what creates value for companies, which is clearly wrong.

[Invited Keynote] - TensorFlow and the AI community

Power the next generation of AI use cases

Machine Learning creates an opportunity to build solutions that address some of the world's most pressing challenges, and deliver positive social impact. In this session, we are going to introduce how Nvidia closely collaborates with developers on the African continent to provide Machine Learning solutions in AgTech, satellite imagery, climate change, etc.

[Invited Keynote] - Why we need a movement to educate 100k ML practitioners across Africa

The opportunity in Africa is to advance the state of the art of applying machine learning to local problems. The underlying rhetoric has been around for decades around the broader application of tech. We’ve seen the benefits of this movement, specifically in the fintech sector, measurable by the impact fintech has on people’s lives. ML offers new kinds of opportunities across various domains but requires a new type of education. I will discuss the options I see to accelerate the trend of local companies and entrepreneurs solving local problems and rethinking higher education to create a globally differentiated workforce in Africa.

Guiding StyleGAN to generate Labeled Synthetic faceimage dataset for under-represented groups

The main cause of bias in AI is the training data. Often, some groups are overrepresented, and others are under-represented. When we have little attention on how data is collected, processed, and organized there will be a lack of geodiversity, which incidentally produces data that leads to gender, ethnic and cultural biases.

ASMSpotter: discover and track artisanal mining using AI

Artisanal small-scale mining (ASM) is a subject of growing importance, especially in the developing world and the African continent. There are an estimated 10 million people working in the sector in sub-saharan countries alone, with numbers growing due to commodity price rises and climate change. The informality and the mining practices do not only cause severe damage to ecosystems and food chain poisoning, but also can lead to social conflicts. These can arise for example between indigenous people and non-local miners or governments and rebel groups financed by ASM. Moreover, ASM sites are known to host child labor, exploitation of workers and human trafficking.

Building a Virtual Cooking Assistant for visually impaired individuals using ML Services on AWS

With Kimessha Paupamah, Saubia Khan & Gabrielle Dompreh.
There are approximately 245 million individuals living with visual impairments, and 300 million diagnosed with some form of colourblindness. One of the major obstacles faced by these individuals is meal preparation. We developed an app using AWS Machine Learning services to assist these individuals prepare food in a safe and healthy manner.

Closing remarks

Machine Learning & Deep Learning for Computer Vision: From Zero to Hero

Fundamentals of Deep Learning Workshop

[Intermediate level]
With Usman Amjad, Muhammad Umair, Twaha Zia & Humera Tariq.
The workshop will go through the basics of how deep learning works through hands-on exercises in computer vision and natural language processing. The attendees of this workshop will train deep learning models from scratch, learning tools and tricks to achieve highly accurate results. They will also learn to leverage freely available, state-of-the-art pre-trained models to save time and get your deep learning application up and running quickly.
Learn to build deep learning and accelerated computing applications for industries such as autonomous vehicles, finance, game development, healthcare, robotics, and more. Obtain hands-on experience with the most widely used, industry-standard software, tools, and frameworks. Gain real-world expertise through content designed in collaboration with industry leaders such as the Children‚ the Hospital of Los Angeles, Mayo Clinic, and PwC. Earn an NVIDIA DLI certificate to demonstrate their subject matter competency and support career growth. Access content anywhere, anytime with a fully configured, GPU-accelerated workstation in the cloud.
The target audiences are students or researchers with:
• Understanding of fundamental programming concepts in Python 3, such as functions, loops, dictionaries, and arrays.
• Familiarity with Pandas‚ data structures
• An understanding of how to compute a regression line.
Please check out the detailed workshop agenda and prerequisites https://developer.nvidia.com/dli/getready

Tidy Modelling in R

[Beginner level]
With Astrid Radermacher, Amieroh Abrahams & Andrew Collier.
The workshop will introduce both the concept of tidy data and tidy modelling in R. This workshop is suitable for those who have never worked in R before, but have some understanding of the model building process.
Participants will be able to :
• wrangle data with the tidyverse suite of packages
• create predictive models using the tidymodels suite of packages.
Prerequisites:
• A laptop
• Internet connection

Fundamentals of Accelerated Data Science with RAPIDS [NVIDIA]

[Intermediate level]
The workshop will teach developers how to build and execute end-to-end GPU accelerated data science workflows that enable to quickly explore, iterate, and productize. Using the RAPIDS data science libraries, developers will be able to use pandas, scikit-learn-like APIs to accelerate a wide variety of GPU-accelerated machine learning algorithms, including XGBoost, K-means, DBSCAN, and logistic regression, etc. to perform data analysis at scale.
•Learn how to implement GPU-accelerated data preparation and feature extraction
•Apply a broad spectrum of GPU-accelerated machine learning tasks and execute GPU-accelerated graph analysis with RAPIDS
•Understand how to leverage GPUs for everyday and advanced data science tasks, and how to use RAPIDS to harness the GPU parallel computing power.
Prerequisites:
•Python programming
•Basic understanding of data science libraries like numpy, pandas, scikit-learn
Please perform the following steps: Review the agenda, prerequisites, and suggested material for the workshop (as detailed in the course datasheet). This is an important step to properly prepare for the workshop. Create or log into your NVIDIA Developer Program account. This account will provide you with access to all of the DLI training materials during and after the workshop. Visit websocketstest.courses.nvidia.com and make sure all three test steps are checked “Yes.” This will test the ability for your system to access and deliver the training contents. If you encounter issues, try updating your browser. Note: Only Chrome and Firefox are supported. Check your bandwidth. 1 Mbps downstream is required and 5 Mbps is recommended. This will ensure consistent streaming of audio/video during the workshop to avoid glitches and delays.

Benchmarking Algorithms for Time Series Modeling

[Intermediate level]
During this workshop, we'll go through the basics of statistical Time Series Modeling that will serve as a baseline of our benchmark. Then we'll go through basic to advanced Deep Learning algorithms and compare their accuracy on a Financial time series based dataset.
After this workshop participants will:
• Acquire knowledge about several Deep Learning algorithms applied to Time Series based data.
• Be able to know which algorithms to use when they need a fast result or highly precise result, and how to combine several different.
Prerequisites
• Basic Python level
• Medium TensorFlow level

Data Science on AWS

[Intermediate level]
With Antje Barth & Chris Fregly.
In this hands-on workshop, we will build an end-to-end AI/ML pipeline for natural language processing with BERT, TensorFlow/Keras and Amazon SageMaker.
Attendees will learn how to
• Ingest data into S3 using Amazon Athena
• Visualize data with pandas, matplotlib on SageMaker notebooks
• Run data bias analysis with SageMaker Clarify
• Perform feature engineering on a raw dataset using Scikit-Learn and SageMaker Processing Jobs
• Store and share features using SageMaker Feature Store
• Train and evaluate a custom BERT model using TensorFlow, Keras, and SageMaker Training Jobs
• Evaluate the model using SageMaker Processing Jobs
• Track model artifacts using Amazon SageMaker ML Lineage Tracking
• Run model bias and explainability analysis with SageMaker Clarify
• Register and version models using SageMaker Model Registry
• Deploy a model to a REST Inference Endpoint using SageMaker Endpoints
• Automate ML workflow steps by building end-to-end model pipelines using SageMaker Pipelines.
Prerequisites:
• Good internet connection and a modern web browser (preferably Chrome).
• Working knowledge of machine learning algorithms and concepts.
• Proficient in Python programming at an intermediate level, and should be familiar with Jupyter notebooks and statistics.
• Preferably attendees also have a basic working knowledge of AWS

Hands-on Multi-Agent Reinforcement Learning using Mava

[Intermediate level]
With Faten Ghriss, Khalil Gorsan Mestiri & Claude Formanek.
This workshop will give participants the opportunity to learn about the theory and practice of multi-agent reinforcement learning (MARL). MARL extends the decision-making capabilities of single-agent reinforcement learning (RL) to the setting of distributed decision-making problems. In MARL, multiple agents are trained to act as individual decision-makers of some larger system, while learning to work as a team. MARL can be applied in situations where the problem becomes exponentially more difficult to solve as it scales. For example, in managing a fleet of autonomous vehicles for a growing population, the number of navigation decisions that must be made at any given time scales exponentially with the size of the fleet. This quickly becomes intractable for single-agent approaches whereas for MARL, is an opportunity to shine. In this workshop, we will show participants how to use Mava, a research framework for MARL to build a multi-agent learning system for a real-world use case. We will provide the necessary guidance, tools and background to understand the key concepts behind MARL, how to use Mava building blocks to build systems and how to train a system from scratch. At the end of the workshop, participants should have a good understanding of key concepts in MARL, potential use-cases and how the computational framework of MARL might differ from other approaches to solving complex problems. On the practical side, participants: Should have the knowledge to be able to frame a real-world problem as a MARL problem and know when this might, or might not, be the appropriate view. Are able to use components and modules from the Mava framework to build a multi-agent system and train this system on a particular example use case.
Prerequisites:
• Participants should be comfortable with programming in Python.
• Have a basic understanding of key concepts in machine learning and in particular, reinforcement learning.
• No prior knowledge of multi-agent reinforcement learning is expected or required.

Deep Learning on ALL the data - beyond the basics with FastAI

[Intermediate level]
If you have RGB images and want to classify them with deep learning, you're in luck - there are more tutorials on that subject than anyone could possibly work through. But what if you have multi-spectral imagery from a satellite, or four-channel audio from a forest microphone array, or maybe songs in some obscure musical notation? In this workshop, we'll be exploring ways to work with more unusual data in ways that let us apply deep learning for more interesting use cases.
Participants will leave with :
• A set of tips and tricks for wrangling unusual data into a useful form
• The ability to build custom data pipelines with Fastai's DataBlock API.
Prerequisites:
• Google account and internet access
• Ideally some experience using Google Colab.

Introduction to Recommendation Systems

[Intermediate level]
The workshop is designed by Google researchers as an introduction to recommendation systems. It covers the basics of recommendation, and a deep dive into some of the commonly used models, including matrix factorization and neural embedding models. A hands-on coding session will allow participants to train and evaluate a movie recommendation system.
• An understanding of how to formulate and train a recommendation problem.
• Hands-on experience training a simple but functional movie recommendation model.
• Experience with inspecting and evaluating recommendation models.
• A self-contained python notebook that can be used as a sandbox for experimenting.
Prerequisites:
• Basic knowledge of machine learning terminology, familiarity with Python for the coding sections.
• Helpful but not required: familiarity with TensorFlow.
• Laptops and internet access. The code/data will be available online.

Moroccan Darija Wikipedia: Basics of Natural Language Processing for a Low-Resource Language

[Beginner level]
With Ihsane Gryech, Khalil Mrini, Abdelhak Mahmoudi, Anass Sedrati & Imane Khaouja.
• The workshop will be divided into two parts: a tutorial where attendees will learn about the basics of Natural Language Processing (NLP), and a practice session where attendees will get to analyze a dataset of Moroccan Darija and present their findings.
• NLP is a field that is in high demand, and where research progresses actively and quickly. Whereas language technology for languages like English and French is highly developed, low-resource languages (like most African indigenous languages) have been left behind and marginalized. There are many opportunities to create new tools for languages with few resources. In this tutorial, we take the example of Moroccan Darija, the national vernacular in Morocco. Our use case dataset will be the Moroccan Darija Wikipedia.
• The participants will first learn statistical tools to analyze language in the tutorial. The tutorial will go over NLP notions including text pre-processing and tokenization, n-gram language modeling, n-gram frequency, topic modeling, and word embeddings. The tutorial consists of theoretical definitions and concrete examples in Python. The participants can then move to the practice part of the workshop, in teams of 1 to 5 people. Each team will be given the Moroccan Darija Wikipedia and will work on analyzing the dataset from an angle of their choice. At the end of the workshop, the teams will be invited to show their findings in a short presentation. Gain basic knowledge of NLP in the tutorial. Practice analyzing text data in the second part of the workshop. Practice data analysis and presentation of results. The workshop is recommended for North African people who aspire to be data scientists, NLP and/or Machine Learning researchers and practitioners, and people interested in computational linguistics. Participants will have the option to get a certificate of participation for this workshop, including the number of hours of participation.
Prerequisites:
• Intermediate familiarity with Python
• No prior knowledge of Natural Language Processing or Machine Learning is required
• Familiarity with any North African Darija is recommended
• Computer with Internet, Python 3, and Jupyter Notebook