7 TILs
**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…
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…
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…
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…
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…
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/2026/08/05/unsup…
[Artificial Analysis](https://artificialanalysis.ai/) is so far the only benchmark that