~/blog/series
Deep Learning
A from-scratch walk through deep learning — perceptrons and backpropagation, activation and loss functions, optimizers, regularization, CNNs, and sequence models like LSTMs and GRUs — each concept worked out on a concrete numeric example.
Deep Learning is the study of models that learn their own features instead of relying on hand-engineered ones. This series builds that idea up piece by piece: starting from a single perceptron, through the mechanics of backpropagation and gradient-based optimization, into the architectures — CNNs for images, LSTMs and GRUs for sequences — that make modern deep learning work in practice.
Every post traces the math on a small, concrete example so you can follow the arithmetic by hand before trusting the code.