A label in machine learning is the ground truth output or annotation for a given data instance. In supervised learning, each training example consists of an input (features) and a label (the desired output). For example, in a labeled image dataset for classification, an image of a dog might have the label “dog.” For regression, labels are continuous values (like house prices). In structured prediction, labels can be complex (like a sequence for each sequence input, or a mask for each image). The term “label” usually implies a human-provided annotation in training data (especially for tasks like image labeling, it’s often a manual process). Good labels are critical – mislabeled data can confuse training. In unsupervised learning, there are no labels (the algorithm tries to find structure without them). In reinforcement learning, “labels” might be in form of rewards which guide learning rather than direct correct outputs.