Posts

Regularization

Data Augmentation, Weight Decay, Dropout, Label Smoothing, and Early Stopping.

Optimizers

Stochastic Gradient Descent, Momentum, RMSProp, and Adam. How to use gradients.

Backpropagation

The first post in the “Notebook of Things I Don’t Know About”. A simple rederivation and NumPy implementation of the Backpropagation algorithm for feedforward networks.

The Many Faces of Mean Field Theory

Variational principles, field theory, and the physics of the Ising Model.

Gaussian Process Regression

Distributions over functions, kernel design, and uncertainty in regression.