Optimization methods that use noisy or approximate gradients instead of exact ones to handle large datasets.