Glossary · Phase 4 · Deep learning
SGD
Gradient descent using a small random batch each step instead of all the data: noisier but much faster.
Learn it properly
Mini-batches and SGD · Train on small random batches instead of everything at once. About 30 minutes.