Speakers - #indabaX

Adele Botha

CSIR, University of Stellenbosch (Dep of Industrial Engineering) & UNISA, (School of Engineering)

AI - The Ethics of IT

Day 3 - 14h30 - 15h30

Healthcare

Arné Schreuder

EPI-USE/University of Pretoria

An Introduction to Spiking Neural Networks and Biologically Plausible Machine Learning

Day 1 - 10h30 - 12h00

Deep Learning

Asad Jeewa

University of KwaZulu-Natal

Fundamentals of Reinforcement Learning

Day 2 - 10h30 - 12h00

Research RL

Asif Aziz

AMBA

Education

Day 2 - 17h30 - 18h00

Education

Athandiwe Saba

Code for Africa

The Power of AI to Fight Information Manipulation

Day 3 - 14h30 - 15h30

AI Ethics AI in Africa Language

Avashlin Moodley

CSIR

Retrieval-Augmented Generation (RAG)

Day 2 - 14h30 - 16h00

Applied ML & Industry

Batsi Ziki

University of Cape Town

Exploring Meta-learned Curiosity Algorithms

Day 1 - 14h30 - 16h00

Research RL

Benjamin Rosman

University of the Witwatersrand

Panelist

Applied ML & Industry

Benjamin Sturgeon

University of Cape Town

An Introduction to AI Safety

Day 2 - 14h30 - 16h00

AI Ethics AI Safety and Governance in Africa Workshop

Bigomokero Antoine Bagula

University of the Western Cape

Transforming Networking Through Artificial Intelligence: From Best Effort to Knowledge Aware Networking

Day 3 - 13h00 - 14h00

AI in Africa MLOps

Talk description

The Internet has evolved into a universal platform for all forms of modern communication. Initially, it was constructed on the principle of an intelligent edge and a core dump, utilizing a best-effort routing model that enabled it to scale massively and support hundreds of millions of nodes. The shift towards a more intelligent Internet began with the introduction of virtualization and the Software-Defined Networking (SDN) paradigm, which successfully separated the control and data planes. A major advancement was achieved with the incorporation of Artificial Intelligence (AI) into networking through the Knowledge-Defined Networking (KDN) paradigm, allowing networks to extract valuable insights from data process Building on this foundation, Knowledge-Aware Networking (KAN) introduces a ground-breaking paradigm that utilizes multi-controller SDN architectures to create innovative learning models for cooperative, competitive, and diverse environments. This approach aims to significantly enhance security, resilience, and intelligence within the network. In KAN, cooperative learning employs the Federated Learning model, where multiple clients collaboratively train machine learning models. Competitive learning involves choosing the most suitable model for a particular use case from a pool of models. Diversity learning replicates the same model across different servers and uses a voting mechanism to ensure fault tolerance. Knowledge-Aware Networking holds the potential to redefine the landscape of networking, embedding intelligence, efficiency, and robustness into the core of our communication infrastructure. This keynote is dedicated to elucidating the transformative impact of artificial intelligence on networking, emphasizing the transition from traditional best-effort approaches to the more sophisticated realm of knowledge-aware networking. Throughout this presentation, we will delve into the intricacies of the KAN architecture, unravelling its layers and exploring the myriad learning models it accommodates. Moreover, we will examine how KAN empowers networks to operate with greater intelligence and autonomy, facilitated by collaborative, competitive, and diversity learning.

Speaker bio

Bigomokero Antoine Bagula, holds a Ph.D. in Communication Systems from the Royal Institute of Technology (KTH) in Stockholm, Sweden. He also obtained two MSc degrees: one in Computer Engineering from the Université Catholique de Louvain (UCL) in Belgium, and another in Computer Science from the University of Stellenbosch (SUN) in South Africa. Presently, Dr. Bagula serves as a full professor in the Department of Computer Science at the University of the Western Cape (UWC) in South Africa, where he leads the Intelligent Systems and Advanced Telecommunication (ISAT) laboratory. Additionally, he holds a professorial position at Université Nouveaux Horizons (UNH) in the Democratic Republic of the Congo (DRC), where he is responsible for driving the institution’s faculty of Informatics research agenda and enhancing its teaching curriculum. Dr. Bagula’s research interests encompass various areas such as Data Engineering with a focus on Big Data Technologies, Cloud/Fog Computing, and Network Softwarization, including concepts like NFV and SDN. He actively explores the potential of the Internet of Things (IoT), covering both the Internet-of-Things and Tactile Internet-of-Things. Moreover, he delves into Data Science, particularly Artificial Intelligence, Machine Learning, and their applications in Big Data Analytics. Dr. Bagula’s expertise extends to Next Generation Networks (NGN), including 5G/6G. Through his academic accomplishments and research pursuits, Dr. Bagula significantly contributes to the fields of computer science, data engineering, and telecommunications.

Bruce Bassett

AIMS

Panelist

Astronomy ML Research

Charl Van Heerden

Saigen

Speech Recognition for Africa

Day 2 - 10h30 - 12h00

AI in Africa Language

Chijioke Okorie

University of Pretoria

Intellectual Property, Data, and Technology, in terms of Governance

Day 3 - 14h30 - 15h30

AI in Africa Legal

Dr Laticha EM Walters

IRD South Africa NPC

AI in Digital Health Governance

Day 3 - 14h30 - 15h30

Healthcare

Edward Khomotso Nkadimeng

NRF-iThemba LABS

Enhancing Dynamic Aperture Estimations in the C70 Cyclotron Using Deep Learning Techniques

Day 2 - 10h30 - 12h00

Applied ML & Industry

Emmanuel Dufourq

African Institute for Mathematical Sciences (AIMS)

Deep Learning and Wildlife Monitoring

Day 1 - 13h00 - 14h00

BioAI Deep Learning Keynote

Frans Meyer

Alphawave

Disciplined Entrepreneurship for AI Startups

Day 3 - 14h30 - 15h30

Applied ML & Industry Workshop

Gabrielle Immelman

MindJoy

Accelerating STEM Education with AI

Day 3 - 13h00 - 14h00

Education

Gloria Pucoe

University of the Witwatersrand and Sol Plaatje University

Utilising Ensemble Cluster for Parkinson's Disease Detection.

Day 2 - 10h30 - 12h00

BioAI Data Science

Heinrich van Deventer

University of Pretoria

Limits of ANNs in Continual Learning

Day 1 - 10h30 - 12h00

Deep Learning

Henrica Makulu Kadira

Medscheme Holdings

Using Data to Inform and Inspire Action

Day 3 - 09h00 - 10h00

AI in Africa Data Science Keynote

Hossein Masoumi Karakani

University of Pretoria

Data Literacy: The New Alphabet of the AI Era

Day 3 - 10h30 - 12h00

Data Science

Talk description

In today’s data-driven world, being literate in data is as fundamental as reading and writing. The rapid growth of data, currently at 2.5 quintillion bytes per day, underpins the operations of organisations and fuels Artificial Intelligence (AI). Data Literacy (DL) has evolved from a mere skill to a basic necessity, critical for navigating and making sense of the vast amounts of data generated every second. This talk will delve into the essence of DL, highlighting its vital importance in the AI era and exploring practical strategies to foster this essential skill across various societal layers, from individual empowerment to systemic educational reforms. By the end of this talk, attendees will understand the critical role of DL in the AI era and be equipped with actionable insights to promote and enhance DL within their organizations and communities. Embracing Data Literacy is not just about staying relevant in a data-rich world; it’s about unlocking the full potential of AI technologies, ensuring ethical use, and making informed decisions that drive progress and innovation.

Speaker bio

Dr. Hossein Masoumi Karakani is a statistician who currently works as a Senior Institutional Researcher at the Department of Institutional Planning, Monitoring, and Evaluation (DIP) at the University of Pretoria (UP). Hossein holds a Ph.D. degree in Mathematical Statistics from UP. He also has a verified certificate in Statistical Learning from Stanford University. In his current role, he is using data to make a positive difference in the lives of students and staff, as well as building interpretable, efficient, ethical, and productionable predictive models that could solve practical challenges in higher education. Prior to joining UP in 2021, he had spent four years in the industry with a primary goal of closing the communication divide that frequently arose within organisations, particularly between data scientists and decision-makers. His primary areas of research focus encompass, among other subjects, Advanced Analytics, Time Series, Statistical Process Control, and Bayesian Statistics. His research findings have been featured in esteemed academic journals, and he has showcased his work at renowned international conferences. In 2023, he was selected to serve as a mentor in the Kaggle BIPOC Mentorship Program, where he assists mentees in enhancing their data science abilities and tackling real-world problems. In addition to his academic and industry work, he won two Silver Medals in Kaggle competitions. Kaggle is the world’s largest data science community and most well-known competition platform for data analytics, with at least one million registered users across the globe. In 2021, Dr. Masoumi Karakani became the first person from UP to enter the global rankings in Kaggle and ranked among the top 1% of data scientists in the world. He holds the belief that data science has the potential to significantly contribute to solving numerous global challenges, and typically, a robust educational foundation is essential for cultivating a profound understanding of the data science domain.

Jabu Mtsweni

Council for Scientific and Industrial Research (CSIR)

The Good and not so Good of Cybersecurity for AI & AI for Cybersecurity

Day 1 - 09h00 - 10h00

AI Ethics Research

Jason Chalom

University of the Witwatersrand, and nCino, Inc (Formally DocFox Inc)

GradSUM a Method to Quantitatively Characterise and Explain Deep Learning Model Behaviour in Several Domains (Unpublished)

Day 1 - 14h30 - 16h00

Vision

Jeroen van Goey

InstaDeep

Day 2 - 16h30 - 18h00

Hackathon

Jessica Rees

Discovery

Retrieval-Augmented Generation (RAG)

Day 2 - 14h30 - 16h00

Applied ML & Industry

Jimoh Saheed Tunji

University Of Kwazulu Nata

Comparative Analysis of Density Functions for Assessing Wind Energy Potential in Africa

Day 2 - 10h30 - 12h00

Applied ML & Industry

Talk description

Title: Comparative Analysis of Wind Speed Probability Distributions for Wind Energy Potential in Nigeria Abstract: This study aims to compare the performance of various probability density functions, including the conventional Weibull, normal, log-normal, and gamma distributions, with the mixture Weibull distribution, in assessing the wind energy potential across different locations in Nigeria. Utilizing 20 years (1995-2015) of wind speed data from 12 meteorological stations—two from each of the six geopolitical zones—the analysis reveals significant findings. Key Findings: • The mixture Weibull distribution demonstrated superior fitting accuracy for wind speed data in most stations, outperforming other distributions based on statistical error metrics such as Kolmogorov-Smirnov and Root Mean Square Error (RMSE). • The mixture Weibull distribution effectively captured the bimodal nature of wind speed distributions in several locations, a feature inadequately represented by the conventional Weibull function. • Northern regions of Nigeria generally exhibited higher wind speeds compared to southern regions, with Lagos displaying the highest average monthly wind speed of 12.71 m/s, and Owerri, also in the southern part, showing the lowest at 2.68 m/s. • The spatial variability of wind resources across Nigeria was highlighted by the mixture Weibull parameters, with the shape parameter (α) ranging from 3.98 to 15.71 and the scale parameter (β) from 3.00 to 12.29 m/s. Conclusion: The findings from this study provide valuable insights for wind energy resource assessment and the development of appropriate wind energy technologies in Nigeria, accounting for the unique wind regimes across different regions. This comprehensive analysis not only advances the understanding of wind energy potential but also informs strategic planning for sustainable energy development in Nigeria.

Speaker bio

Saheed is a distinguished Radiophysicist turned Renewable energy expert due to his dream for Africa with sufficient energy supply. He is currently PhD research student at the University of KwaZulu-Natal (UKZN) in South Africa. His research focuses on the “Comparative Analysis of Density Functions for Assessing Wind Energy Potential in Africa”. He started from Nigeria and extending his research to other Africa Countries in due time. This groundbreaking work leverages the power of the R and Python programming languages to enhance the accuracy and reliability of wind energy assessments. With a deep passion for renewable energy and a strong foundation in statistical analysis, Saheed is at the forefront of integrating machine learning with physics and renewable energy development in Africa. His innovative approach aims to identify optimal locations for wind energy generation by analyzing wind speed data using advanced probabilistic models, including the bimodal mixture Weibull function. Saheed’s research holds significant promise for advancing sustainable energy solutions in Africa, contributing to the region’s energy independence and environmental sustainability. By incorporating state-of-the-art machine learning techniques, his work not only addresses current challenges in wind energy potential assessment but also paves the way for future advancements in the field. Through his dedication and expertise, Saheed is poised to make a lasting impact on the renewable energy landscape, demonstrating the potential of interdisciplinary approaches to solving critical energy challenges in Africa and beyond.

Jocelyn Mazarura

Discovery

Introduction to Natural Language Processing

Day 1 - 14h30 - 16h00

Deep Learning Language

Joel Leonard

University of Pretoria

Application of A Discrete Event- Based Monte Carlo Simulation for Improving Reliability, Availability and Maintainability (RAM) Of an SO2 Abatement Plant

Day 2 - 10h30 - 12h00

Applied ML & Industry

Talk description

The modern mining industrial sector greatly depends on an efficient maintenance plan to achieve maximum failure prevention, and a plethora of innovative solutions have been developed to enhance and optimize this endeavor. However, newer plants with complex technologies are built every day and the experience or failure history needed to predict equipment performance, reliability, availability, and maintainability (RAM) especially at the infant stage of their operation is predominantly scarce. To curtail this challenge, an early life RAM study can be adopted, serving a proactive maintenance strategy implemented to optimize the overall lifespan of the system by addressing and resolving early failures and potential future threats. In this research, a discrete event(failure)-based Monte Carlo simulation is used to model a Sulphur dioxide (SO2) abatement plant as a case study in South Africa. The result of the simulation predicts the worst case and best-case scenario of the plant’s condition, number of failures, downtime contributors, reliability and availability for the next 6 years. Furthermore, reliability centered maintenance is implemented on the most critical identified subcomponents and a task list is developed and packaged to form part of the maintenance plan needed to improve the SO2 abatement plant’s health condition. The research shows that numerous complex maintenance challenges of various plants including mining, oil and gas, aviation etc. can be solved using an effective simulation tool which leads to high business productivity, safety and profitability.

Speaker bio

Joel Leonard is researcher at the Centre for Asset Integrity and Management, department of Mechanical and Aeronautical Engineering at the University of Pretoria and a Lecturer at Terotechnica Asset Management College where he teaches Reliability Engineering. He is a member of the SA Institution of Mechanical Engineering (SAIMechE) and the Institution of Engineering and Technology (IET), staying up to date with the latest developments in the industry, especially in the field of Physical Asset Management. As an engineer, Joel has experience in developing and implementing effective strategies for maintaining and improving the reliability of critical equipment. In addition to his professional pursuits, Joel has a keen interest in grassroot education activism, and enjoys travelling, hiking, and trying new delicacies. He is a dedicated individual with a strong work ethic and a commitment to excellence in all aspects of life.

Jonathan Shock

University of Cape Town

Reinforcement Learning: Meta, Offline, Multi-agent and more

Day 1 - 14h30 - 16h00

Research RL

Karabo Madibela

University of the Western Cape

Decoding the Brain: The Role of AI in Neurodevelopmental Health Optimization

Day 1 - 10h30 - 12h00

Applied ML & Industry

Kayode Adetunji

MADIVA Research Hub

Leveraging AI and ML for Innovations in African Healthcare.

Day 3 - 14h30 - 15h30

Explainable AI Healthcare

Kevin Colville

NICIS CHPC (CSIR)

Centre for High Performance Computing

Day 1 - 17h30 - 18h00

Applied ML & Industry Research

Kgomotso Moretsele

WeThinkCode_

Insights on Preparing for the Workplace

Day 2 - 14h30 - 16h00

Education Workshop

Kirsten Taneall Smith

Opti-Num Solutions

Design and Deployment of AI for Innovation in Medical Imaging, Leveraging MATLAB and C++

Day 2 - 16h30 - 17h00

Applied ML & Industry Healthcare

Leo John Hyams

AI Safety Cape Town

An Introduction to AI Safety

Day 2 - 14h30 - 16h00

AI Ethics AI Safety and Governance in Africa Workshop

Mandlenkosi Zwane

Introduction and Demonstration of AI-Based Segmentation for Medical Images

Day 1 - 14h30 - 16h00

Applied ML & Industry

Mayur Ranchod

Sybrin

Is Seeing Really Believing Anymore? 👀

Day 1 - 14h30 - 16h00

Vision

Michelle Skelton

UCT

DS-I Africa - Harnessing Data Science for Health Discovery and Innovation in Africa

Day 3 - 14h30 - 15h30

Healthcare Research

Talk description

Overview of the network, collaborations and outputs.

Speaker bio

Dr. Michelle Skelton: Pan-African Research Coordination in biomedical and data sciences Dr. Michelle Skelton is a seasoned professional managing the H3Africa and DS-I Africa Administrative Coordinating Centre (H3ACC & DS-I Africa CC) with a wealth of experience in project management, coordination, fiscal management, and communications. She excels in fostering transparency, fairness, and timely exchange of critical information while balancing the diverse needs of multiple stakeholders within budget constraints. Dr. Skelton plays a pivotal role in the adaptability and pressure-handling capabilities of the coordinating center, ensuring seamless operations as the consortium’s requirements evolve. Key Responsibilities and Achievements Manages and organizes activities for the Steering Committee, executing joint decisions effectively. Served as DS-I Africa SC Chair. Instrumental in establishing the H3Africa DBAC committee, coordinating with data and biorepository teams. Facilitates strategic communication within the H3Africa and DS-I Africa network, coordinating virtual and face-to-face meetings with NIH, Wellcome and SFA collaboration. Actively engages in outreach and sustainability efforts, securing support from influential organizations and advocating for African science and research sustainability. Successfully facilitated high-profile engagements with dignitaries like Her Excellency Dr. Ameenah Gurib-Fakim and other esteemed guests at consortium meetings. Leads the H3Africa Sustainability efforts, by targeting regional and global organizations to strengthen scientific connections across Africa and funding applications Initiates collaborations with pan-African networks like AfSHG, Africa Pharmacogenomics Network (APN), Data Science Africa (DSA), Deep Learning Indaba/Indaba X, fostering partnerships for research and development initiatives. Drives career development through the H3Africa and DS-I Africa Early Career Researches Initiatives, administering student travel fellowships and organizing workshops on critical skills development. Dr. Skelton’s dedication to advancing African science, fostering collaborations, and promoting sustainability in medical health research and data science exemplifies her commitment to the African Health agenda. Her multifaceted role in coordinating key activities, engaging stakeholders, and driving strategic initiatives underscores her invaluable contributions to the field of genomics and data science and pan-African research coordination.

Mixo Ngoveni

Red Bull Basement University - Empowering Young Minds to use Technology for Positive Impact

Day 1 - 16h30 - 17h00

Education Social

Mortda Adam

University of KwaZulu-Natal (UKZN)

Advancements in Vehicle Detection for Autonomous Drivin

Day 1 - 14h30 - 16h00

Vision

Mpho Dipitso

WeThinkCode_

Insights on preparing for the workplace

Day 2 - 14h30 - 16h00

Education Workshop

Mpho Raborife

Institute for Intelligent Systems, Primus Tech Hub

NLP for Africa: Trends and Perspectives

Day 2 - 09h00 - 10h00

AI in Africa Keynote Language

Neelan Pather

Analytix Engine

The Antidote to ML Hype

Day 1 - 10h30 - 12h00

Applied ML & Industry ML

Pelonomi Moiloa

Lelapa AI

Organizing African Talent to Move Humanity Forward: Language Technology for Africa

Day 0 - 13h30 - 14h30

AI in Africa Keynote Language

Peter Ochieng

University of Cambridge

Exploiting Phonological Similarities between African Languages to achieve Speech to Speech Translation

Day 2 - 10h30 - 12h00

Data Science

Philip Sterne

Minerva University

Teaching Computer Science in the age of LLMs

Day 2 - 10h30 - 12h00

Applied ML & Industry

Privolin Naidoo

CSIR

Leveraging RAG to Improve LLM Generation

Day 2 - 14h30 - 16h00

Applied ML & Industry

Shane Acton

Instadeep

Exploring SOTA methods in BioAi

Day 2 - 17h00 - 17h30

BioAI Research

Simon Rosen

University of the Witwatersrand

Optimal Task Generalisation in Multi-Agent Reinforcement Learning

Day 1 - 14h30 - 16h00

Research RL

Talk description

In this talk I will present our workshop paper “Optimal Task Generalisation in Multi-Agent Reinforcement Learning” which will be appearing at the CoCoMARL workshop being held at RLC 2024. In this paper we present a method for provably optimal task generalisation in multi-agent reinforcement learning, which to the best of our knowledge is the first work to do so. Multi-agent systems where multiple agents must collaborate to achieve multiple tasks are ubiquitous in real world applications. A notable example of this is automated warehouse where multiple autonomous robots must navigate to, and transport assigned shelves. In many situations there are common properties across tasks leading to the possibility of shared knowledge, such as in automated warehouses where the layout of the warehouse and robots do not change but the desired shelves do. In this work we consider goal-oriented tasks where each agent has a set of goals (terminal states) they wish to reach, and each goal has an associated terminal reward. We propose a method Goal-Oriented Learning for Multi-Agent Reinforcement Learning (GOLeMM) that explicitly learns a goal-oriented value function that provably encodes how each agent can optimally achieve every goal. This goal-oriented value function is learnt from a single arbitrary task and afterwards this information is used to optimally infer without further learning any other task in the distribution given only knowledge of the terminal rewards. Empirically we demonstrate that given a fixed step quota our method is able to optimally generalise over a full task distribution while representative methods can only learn a small fraction with the gap between our method and those increasing exponentially as the number of goals in the environment increases.

Speaker bio

I am a PhD candidate in Computer Science at the University of the Witwatersrand and a part of the RAIL Lab. Cooperative multi-agent problems are ubiquitous in real-world applications, and cooperative multi-agent reinforcement learning is a popular solution to this class of problems. MARL is hampered by a myriad of problems such as poor sample efficiency and poor agent scaling. In my work I wish to address these problems in a mathematically grounded manner along with rigorous empirical experiments. I believe that MARL has the possibility to revolutionise entire industries and improve the world for the better and I believe my research is taking important steps towards that goal.

Siphiwe Thwala

IBM Research

Biomedical Text Insights Extraction

Day 3 - 14h30 - 15h30

BioAI Data Science

Steven James

University of the Witwatersrand

Concurrent and Temporal Composition for Zero-shot Transfer in Reinforcement Learning

Day 1 - 14h30 - 16h00

Research RL

Sthembiso Mkhwanazi

CSIR and NWU

Breaking Down isiZulu texts: Supervised Segmentation for Better Language Modeling and Translation

Day 2 - 14h30 - 16h00

Applied ML & Industry

Takura Wekwete

Rand Mutual Assurance / University of Pretoria

Reinforcement Learning in Financial Risk Management

Day 1 - 10h30 - 12h00

Research RL

Tam Ogubai

Zindi

It Takes a Ecosystem for Data Science to Succeed.

Day 1 - 17h00 - 17h30

Applied ML & Industry Data Science

Tariq Blecher

South African Radio Astronomical Observatory

ML Approaches to Radio Interferometric Imaging

Day 1 - 14h30 - 16h00

Vision

Tawanda Ewing

Legal Interact

Bridging the Gap to AGI - The Thinking Graph

Day 1 - 10h30 - 12h00

Applied ML & Industry

Terence Van Zyl

University of Johannesburg

Deep Learning : The Journey from Feature Engineering to Representation Learning

Day 1 - 10h30 - 12h00

Deep Learning Foundations

Tsapang Mashego

Enhancing Autonomous Driving: An In-depth Analysis of Deep Learning-based Object Detection Techniques

Day 1 - 14h30 - 16h00

Deep Learning

Verosha Pillay

Melio.ai

Exploring MLOps with Melio AI and Zindi

Day 1 - 10h30 - 12h00

Applied ML & Industry

Zubeida Dawood

CSIR/ TUT/ WITS Business School

Real-world Applications of Ontology Engineering in Deep Learning : A Cybersecurity Learning Factory Use Case

Day 3 - 13h00 - 14h00

Deep Learning Research