Machine Learning - csen2025.org

Dr. Ingrid Falk-Weller
September 8, 2026
(0)
Machine Learning

Adam Converges Faster and Generalises Differently

I remember sitting in a windowless lab three years ago, staring at a loss curve that looked more like a mountain range than a smooth descent. I had swapped out my standard Adam implementation for a “state-of-the-art” variant I’d read…

Deep training made practical via batch normalisation.

Dr. Ingrid Falk-Weller
September 3, 2026
(0)
Machine Learning

Normalising Between Layers Made Deep Training Practical

I remember sitting in a windowless lab three years ago, watching a training loss curve oscillate so violently it looked more like a seismograph reading than a convergence plot. I had followed every “best practice” in the literature, yet my…

Deep networks facing vanishing and exploding gradients.

Deep Networks Fail to Train Before They Fail to Generalise

I remember sitting in a windowless lab during my PhD, staring at a loss curve that looked less like a descent and more like a flatline on a heart monitor. I had spent three weeks tuning hyperparameters, convinced I was…

Activation functions compared in neural networks.

Without a Nonlinearity the Whole Network Collapses to One Layer

I remember sitting in a windowless lab three years ago, watching a training loss curve flatten into a perfectly straight, useless line. I had spent forty-eight hours tuning hyperparameters, only to realize I had blindly defaulted to a Sigmoid function…

Diagram of backpropagation explained via chain rule.

The Chain Rule Applied Very Carefully and Very Often

I spent three years in academia watching brilliant students stare at chain-rule derivations until their eyes glazed over, all because they were taught that backpropagation is some mystical, impenetrable sorcery. I’ve seen too many tutorials treat it like a black…

Visualizing neural network fundamentals through functions.

A Stack of Simple Functions That Becomes a Complicated One

I spent three years in academia watching brilliant researchers build increasingly baroque mathematical proofs for things that could be explained with a simple diagram and a bit of intuition. It frustrates me how often people treat neural network fundamentals as…

Implementing class imbalance strategies via resampling.

Resampling Fixes the Metric and Sometimes Nothing Else

I remember sitting in a windowless server room during my first industry role, staring at a training log that boasted 99.4% accuracy while the model failed to catch a single actual fraud case. It was a gut punch. We had…

Calibration of probabilities in predictive modeling.

A Confident Model Is Not Necessarily a Correct One

I remember sitting in a windowless conference room three years ago, watching a lead researcher present a model that boasted a near-perfect accuracy score, only to watch our entire deployment strategy crumble because the system was fundamentally overconfident. We had…

Auc Measures Ranking, Not Calibration

I remember sitting in a windowless lab during my postdoc, staring at a training log that claimed a near-perfect 0.99 AUC for a fraud detection model. My supervisor was ready to celebrate, but I couldn’t shake the feeling that something…

Graph illustrating the precision recall tradeoff.

Moving the Threshold Trades One Error for the Other

I remember sitting in a windowless lab during my postdoc, staring at a training curve that looked perfect on paper but was absolute garbage in practice. My supervisor was celebrating a near-perfect accuracy score, but I was watching the logs…