Stochastic Gradient Descent (SGD) is an optimization algorithm used to minimize a function, commonly in machine learning. Unlike traditional gradient descent, which calculates the gradient using the entire dataset, SGD updates the model's parameters using only one or a few training examples at a time. This makes it faster and more efficient, especially for large datasets. SGD helps in finding the best parameters for models like neural networks by iteratively adjusting them based on the error of predictions. Although it can introduce noise in the updates, this randomness often helps escape local minima, leading to better overall solutions.