Stochastic gradient descent and Adam are optimization algorithms that update model parameters from estimated gradients, but they use different rules for momentum and per-parameter step sizes. This guide explains the mechanism, trade-offs, evaluation, and controls that matter in practice.