Machine Learning - csen2025.org

Comparison of categorical encoding methods.

High Cardinality Breaks One Hot Encoding Quietly

I spent three months in a research lab during my PhD chasing a marginal accuracy gain, only to realize I had spent half that time debugging a catastrophic data leak caused by a poorly implemented one-hot encoder. It is infuriating…

Analyzing the impact of handling missing data.

Why the Data Is Missing Matters More Than That It Is

I remember sitting in a windowless lab during my PhD, staring at a dataset that looked less like a coherent signal and more like a sieve. I had spent three weeks trying to implement a sophisticated, multi-stage Bayesian imputation model…

Learning feature engineering fundamentals for model training.

Dr. Ingrid Falk-Weller
February 28, 2026
(0)
Machine Learning

The Model Learns What You Show It, Nothing More

I remember sitting in a windowless lab during my postdoc, staring at a loss curve that refused to budge, no matter how many layers of transformer blocks I stacked onto the architecture. I had spent weeks optimizing hyperparameters and chasing…

Diagram illustrating the bias variance tradeoff.

Dr. Ingrid Falk-Weller
February 22, 2026
(0)
Machine Learning

Underfitting and Overfitting Are the Same Dial

I remember sitting in a windowless lab three years ago, staring at a model that had achieved near-perfect accuracy on my training set, only to watch it fall apart the moment I fed it real-world data. I had fallen into…

Cross validation explained through multiple estimates.

Dr. Ingrid Falk-Weller
February 10, 2026
(0)
Machine Learning

Five Estimates Beat One Lucky Split

I remember sitting in a windowless lab during my PhD, staring at a training curve that looked absolutely perfect—too perfect. I had spent weeks tuning a model, feeling that intoxicating rush of seeing near-zero error rates, only to have the…

Applying regularisation techniques to improve model generalisation.

Making the Model Worse on Purpose So It Generalises

I spent three years in academia watching researchers treat regularization techniques like some kind of dark magic—a collection of opaque hyperparameters you just “tune” until the loss curve looks pretty. I remember sitting in a windowless lab, staring at a…

Visualizing overfitting and how to see it.

A Model That Memorises Looks Perfect Until It Meets Reality

I remember sitting in a windowless lab three years ago, staring at a loss curve that looked like a work of art—a perfect, smooth descent toward zero. My training metrics were flawless, and I felt that rush of dopamine we…

Visualizing train validation and test splits.

Touching the Test Set Once Is Once Too Often

I still remember the exact moment I realized my PhD thesis was built on a house of cards. I was staring at a convergence plot in a dimly lit lab, feeling that heady rush of success, only to realize that…

Supervised versus unsupervised learning cost comparison.

Labels Are Expensive, Which Shapes the Whole Field

I spent three years in academia watching brilliant researchers build massive, expensive models for supervised versus unsupervised learning tasks, only to see those same models crumble the moment they hit a messy, real-world dataset. The industry loves to sell these…