Simon Valentin’s Post

This machine learning model was trained on a large multicenter dataset of unruptured intracranial aneurysms to predict future rupture risk across international cohorts, supporting more precise and personalized treatment decisions. https://ja.ma/4jfqSJC

Recently worked on a CNN-based Pneumonia Detection system 🩺🤖 Here the real learning wasn’t just building a model, but understanding how ML systems behave under uncertainty. While the network achieved ~86% test accuracy with a ~96% recall (critical for screening tasks), the deeper work was analysing why confident false positives occur, why some true cases are missed, and how prediction confidence should be interpreted—not blindly trusted. By combining Grad-CAM with TP / FP / FN analysis, I learned how classification models rely on global texture patterns rather than precise locations, and why explainability is about evidence, not diagnosis. This project reshaped how I evaluate ML models—less focus on metrics alone, more focus on decision reliability and limitations. #DeepLearning #MachineLearning #CNN

ACM Transactions on Probabilistic Machine Learning (TOPML) is pleased to introduce three new continuous special sections to give ongoing visibility to areas where probabilistic machine learning is evolving rapidly. Visit TOPML website for more information and submission instructions. https://lnkd.in/eewnHjqS @acm_tompl

K-Nearest Neighbors looks almost too simple yet it’s used in recommendation systems, anomaly detection, and many real-world ML workflows. In this post from my #MLRefreshSeries, I’ve covered: 👉 What KNN really is 👉 How the choice of K and distance metrics drives predictions 👉 Why feature scaling is critical for KNN 💬 What’s one problem you’d try KNN on and what distance metric would you choose? #MachineLearning #DataScience #ML #MLRefreshSeries

The Impact of Semantics on Machine Learning: Unlocking the Power of Meaning Have you ever wondered why your smart assistant sometimes misunderstands you? It all boils down to semantics—the study of meaning that plays a critical role in how machines interpret language. In this article, we’ll explore the fascinating relationship between semantics and machine learning, highlighting its significance in creating more intelligent systems. Get ready to discover … 𝗔𝗱𝗱 𝘃𝗮𝗹𝘂𝗲 𝘁𝗼 𝘆𝗼𝘂𝗿 𝗟𝗶𝗻𝗸𝗲𝗱𝗜𝗻. 𝗚𝗶𝘃𝗲 𝘂𝘀 𝗮 𝗳𝗼𝗹𝗹𝗼𝘄! @Englezz

The Impact of Semantics on Machine Learning: Unlocking the Power of Meaning Have you ever wondered why your smart assistant sometimes misunderstands you? It all boils down to semantics—the study of meaning that plays a critical role in how machines interpret language. In this article, we’ll explore the fascinating relationship between semantics and machine learning, highlighting its significance in creating more intelligent systems. Get ready to discover … 𝗔𝗱𝗱 𝘃𝗮𝗹𝘂𝗲 𝘁𝗼 𝘆𝗼𝘂𝗿 𝗟𝗶𝗻𝗸𝗲𝗱𝗜𝗻. 𝗚𝗶𝘃𝗲 𝘂𝘀 𝗮 𝗳𝗼𝗹𝗹𝗼𝘄! @Englezz

📢 Revolutionary Model Transforms Experiment Design Efficiency! 🤖 Researchers from Imperial College London and DataHow have developed a new intelligent sampling algorithm. 📊 This model enhances the Design of Experiment DoE process for upstream activities. ✅ Unlike traditional methods, it can handle multiple variables simultaneously. 🔍 The aim is to make machine learning tools accessible for better industrial insights. 🛠️ The algorithm is currently in use and being tested for various applications. #BioprocessUpdates #DesignOfExperiment #IntelligentModel #MachineLearning #DataScience #ResearchInnovation #AIinIndustry #UpstreamActivities #SamplingAlgorithm #IndustrialInsights #DataHow #ImperialCollegeLondon #ExperimentalDesign #TechForGood #ResearchAndDevelopment #SmartAlgorithms ▷ Read the full article here: 📎 https://lnkd.in/ewedvpqR

I recently read the machine learning research paper "A Few Useful Things to Know about Machine Learning" by Pedro Domingos, and one statement particularly stand out for me. #MachinLearning #ResearchPapers

Recently explored hyperparameter tuning in machine learning using Logistic Regression. Started with a baseline model that achieved 96.49% accuracy, then applied #GridSearchCV with cross-validation to tune hyperparameters like regularization strength. After tuning, the model reached 98.25% accuracy on unseen test data, with better generalization. #Result: Error reduction: The baseline model made 4 mispredictions, while the tuned model made only 2. This experiment reinforced an important lesson for me: #Hyperparameter tuning isn’t just about boosting accuracy, it’s about guiding how a model learns. Learning a lot by building and experimenting step by step. More ML work coming. #MachineLearning #HyperparameterTuning

𝗣𝗲𝗿𝗰𝗲𝗽𝘁𝗶𝗼𝗻 𝗟𝗲𝗮𝗿𝗻𝗶𝗻𝗴 (𝗣𝗲𝗟) 𝗶𝗻 𝗮 𝗯𝗿𝗼𝗮𝗱𝗲𝗿 𝗔𝗚𝗜 𝘀𝘁𝗮𝗰𝗸. Perceptual state (Z) supports world modelling, planning/decision learning, retrieval/knowledge, and task heads. Task/return gradients do not flow into the sealed/fire-walled perception encoder (f_phi).