Blog
Blog by Arthur Cartel Foahom Gouabou - Articles on AI, Machine Learning and MLOps
Articles
Build a RAG chatbot: a practical step-by-step guide
March 24, 2025
A practical tutorial using French fuel-price data, retrieval-augmented generation and a complete GitHub project.
Step-by-Step Guide: Monitoring ML Model Performance with Prometheus & Grafana
March 5, 2025
In this hands-on tutorial, I walk you through deploying a Streamlit-based sentiment analysis app, and setting up Prometheus and Grafana to monitor real-time model performance. Learn how to track metrics such as accuracy and number of requests, using Docker and Python for a fully containerized monitoring solution.
Bringing AI to Real-World Healthcare: Key Challenges and Solutions
November 15, 2024
In this article, I explore the challenges of implementing AI in real-world healthcare environments, from adapting to diverse patient populations to ensuring fairness and explainability in AI decision-making. The piece highlights practical solutions such as external validation, training on diverse datasets, and leveraging explainable AI tools to enhance trust and utility. By tackling these issues, AI can become a reliable partner in patient care, improving outcomes while maintaining fairness and transparency.
Enhancing Medical Predictions: A Comprehensive Guide to Model Calibration
July 23, 2024
In this article, I explore the crucial role of model calibration in medical predictions, emphasizing its importance for accurate risk assessments in diverse patient populations. The article discusses key calibration metrics, such as the Brier score, calibration plots, and Expected Calibration Error (ECE), highlighting their strengths and limitations. It also presents practical methods like Platt Scaling for enhancing model calibration, accompanied by a step-by-step Python implementation for classifying chest X-ray images. This comprehensive guide aims to improve predictive models' reliability, ultimately leading to better patient outcomes.
Understanding Confidence Intervals in Medical Task Prediction
June 24, 2024
In this article, I delve into the concept of confidence intervals, a critical statistical tool often encountered in medical data science. Drawing from my experience during a master's internship at Aix-Marseille University, where I worked on melanoma diagnosis, I aim to demystify confidence intervals for novice data scientists. The article explores the importance of confidence intervals in estimating the performance of predictive models, highlighting their role in conveying uncertainty and ensuring the reliability of model predictions. By understanding and correctly interpreting confidence intervals, clinicians and data scientists can make more informed decisions, ultimately enhancing patient care and trust in medical applications of AI.
Federated Learning and AI in Medicine: Protecting Data, Improving Care
May 14, 2024
Artificial intelligence (AI) and machine learning have transformed the healthcare sector by improving the accuracy of diagnoses and treatments. However, the increased use of sensitive data raises significant concerns about privacy and security. Federated Learning (FL) offers an innovative solution by decentralizing model training. Instead of centralizing data, FL allows it to be kept locally, sharing only model updates and thereby reducing the risk of data breaches. This process is particularly crucial in medicine, where protecting patient data is paramount and regulated by laws such as GDPR and HIPAA. This article explores the principles of Federated Learning, illustrating its application in medical image analysis. We will examine the benefits of this technology, the associated technical challenges, and potential solutions. Finally, we will consider the future prospects of FL and its potential impact on medical practices and beyond.
Mastering Long-Tailed Distribution in Deep Learning for Image Classification: Solutions Unveiled
November 13, 2023
Deep learning, a powerful tool in modern AI, faces a significant challenge when dealing with unevenly distributed data—a phenomenon known as long-tailed distribution. This article explores the impact of long-tailed distribution on image classification models, explaining why it matters in deep learning. We delve into what long-tailed distribution is, how it disrupts AI models, and the strategies experts are employing to tackle this issue. By examining a public dataset of skin lesions, we highlight the disparity between common and rare lesions, illustrating the concept of "Head" and "Tail" in data distribution. Understanding this issue is crucial for improving the performance and accuracy of deep learning models.
How Computers Can Aid Skin Cancer Diagnosis: An Explainer
October 20, 2023
Welcome to a world where computers collaborate with doctors to enhance skin cancer detection. This article demystifies the science behind it, using clear visuals for easy comprehension. Based on a scientific paper, we explore the growing challenges in early melanoma detection due to the rising number of cases and a shortage of dermatologists. We delve into the role of Convolutional Neural Networks (CNNs) in melanoma detection, likening their meticulous image analysis to that of skilled investigators. However, neural networks often appear as "black boxes," and we explain how game theory brings transparency to their decision-making process. Interested readers can dive deeper into the full article and access the open-source research code.