Chris Fourie#
I am a medical doctor turned health data scientist, trying to use the powers of AI for social good.
I currently work full-time as a research and machine learning engineer for analysis of continuous monitoring health data at LifeQ. Part-time I provide consulting services for digital health, artificial intelligence (AI) / machine learning (ML) and software engineering. I co-founded SisonkeBiotik, a grassroots participatory research community making AI/ML for health research in Africa more accessible. I also co-founded Grassroots-Parti, a collective of communities with a focus on helping incubate and provide mutual support for grassroots participatory communities in Africa.
I have a medical degree (MBBCh) and a master's in computer science from the University of the Witwatersrand in South Africa. I carried out master's my research under the supervision of Prof. Benjamin Rosman as a member of the Robotics, Autonomous Intelligence and Learning Lab.
I am motivated by my experience growing up in South Africa, witnessing the devastating effects of perpetuated inequality. During my medical training, I saw first hand the challenges this has created and the impact that resource scarcity can have on the health and wellbeing of individuals. My long term research and career interest intersect at collaboratively developing solutions to address inequality, with consideration for the diverse and complex nature of our societal systems.
Curriculum vitae#
Research#
Fundamental research interests:
- Collective intelligence
- Machine learning with a focus on reinforcement learning
- Computational neuroscience
- Multi-agent modelling
- CoAgent networks
- Emergent communication
- Active inference
Applied research interests:
- Responsible AI
- MLOps / computational research ops / end-to-end computional research automation
- Rapid model benchmarking pipelines
- Meta-learning, hyperparameter tuning
- Human-in-the-loop learning
- AI for healthcare
- Reinforcement learning for data driven public health policy and dynamic treatment regimes
- Privacy preserving AI for health
- Distributed and experimental governance
Communities and academic affiliations#
- SisokeBiotik (co-founder) - [community site], [YouTube], [LinkedIn], [Twitter]
- Grassroots Parti (co-founder) - [community site]
- RAIL Lab (member) - [blog], [members]
- Deep Learning Indaba (community organizer) - [community site]
- IndabaX South Africa (community organizer) - [community site]
- Computational Neuroscience Imbizo (alumni) - [summer school]
Industry#
Independent contractor for LifeQ
- Research engineer
- MLOps / computational research ops
- Time series health data analysis
- Remote and continuous health monitoring
- IoT for health
Independent contractor for RetroRabbit
- Conducting software engineering technical interviews
Independent digital health consultant
- Health data engineering
- Privacy preserving AI for health data
Selected publications#
- "A Framework for Grassroots Research Collaboration in Machine Learning and Global Health", ML for Global health workshop ICLR 2023 [paper]
- Chapter Co-Author on "Trauma and Injury" for Essentials of Global Health textbook (British Medical Journal Book Awards, 1st place in Public Health division, published by Elsevier, edited by Parveen Kumar) - [textbook]
Selected ComSci MSc work#
Research report#
Impact of Noise on Learned Value Functions at Depth in CoAgent Networks for Neural Network Credit Assignment
CoAgent networks (CoANs), networks of reinforcement learning agents, have been shown to be a biologically plausible alternative to backpropagation for solving the neural network structural credit assignment problem [Gupta et al. 2021]. This is accomplished where many agents, each as a neuron in a stochastic neural network, use only their local policy gradient and a global reward. Noise is an important consideration in the learning dynamics of any stochastic neural network [Schoenholz et al. 2017]. We investigate the impact of noise on learnt value functions for baselines and Actor-Critic methods in CoANs at depth. We demonstrate that with additional layers, CoANs using REINFORCE, REINFORCE with a baseline or Actor-Critic methods perform significantly worse. However unbiased variance reduction methods are effective at alleviating this to a moderate extent. For CoANs of increasing depth and width using Actor-Critic methods we show that learned value functions are more sensitive to noise. We show as well that large bootstrapping bias impacts Actor-Critic CoAN methods significantly.
Projects#
Machine Learning#
- Semantic Segmentation for Histopathology (Python, PyTorch) -[notebook][repository][report]
Reinforcement Learning#
- Temporal Difference Learning (Python) - [repository]
- Dynamic Programming (Python) - [repository]
- k-armed bandits (Python) - [repository]
Computer Vision#
- GMatchNet, Graph Neural Networks for Feature Matching - [repository], [video], [report]
High Performance Computing#
- Distributed Deep Learning (C, MPI, CUDA) - [repository] [report]
- Parallel image convolutions (C, CUDA) [repository]
- Parallel kNN and Search (C, OpenMP) [repository]
Algorithm Analysis#
- Recursive Algorithms - Find max sub-array (Java) - [repository]
- Search trees - Binary, Red Black and Order statistic (Java) - [repository]
Robotics#
- SLAM with Kuri (Simultaneous Localisation and Mapping) (Python, ROS) - [repository], [video]
- Object Detection (Python, ROS) - [repository]
- Quadcopter, PID controller (Python, ROS) - [repository]
- Turtlebot (Python, ROS) - [repository]
Thanks for stopping by (",)