#machine-learning

6 items tagged “machine-learning”

TILs (6)

What is Keras, and how is a single neuron different from OLS?

**TL;DR:** Keras is the high-level API for building and training neural networks — not a universal machine-learning API. Regression is a *task type*, not a model kind: the same Keras machinery does it…

Why cross-entropy, and what −1/m is really doing

While building a single neuron from scratch in NumPy for the DLH machine-learning curriculum (classifying MNIST handwritten digits), I had to answer the question every ML beginner meets: how do you sc…

Gradient descent is just walking downhill — and why (A − Y) is enough

The DLH machine-learning curriculum had me build a single neuron from scratch in NumPy — classifying MNIST handwritten digits, the same project as my [training loop](/til/ai/training-a-single-neuron-l…

A neural network is your single neuron, stacked twice — the shapes tell the story

After building a single neuron from scratch in NumPy for the DLH machine-learning curriculum, the next step was the jump that gives "deep" learning its name: stacking two layers instead of one. We wer…

Training a neuron is a two-step heartbeat: predict, then correct

In the DLH machine-learning curriculum I built a `Neuron` class in NumPy that classifies MNIST digits: it takes an image's pixels as inputs **X**, weighs them (**W**), adds a bias (**b**), and outputs…

EM: the algorithm that alternates guesses

While building a Gaussian Mixture Model from scratch in NumPy for the DLH machine-learning curriculum (the clustering module — same project as my [unsupervised clustering notes](/blog/2026/07/28/unsup…