What is K-Shot?

K-shot is a machine learning terminology where k represents the number of labeled examples available per class during training or adaptation, defining the data constraint under which a model must learn new tasks. The variable k typically ranges from 0 (zero-shot) to small integers like 5-10 (few-shot), with higher values indicating more available training examples. This notation originated from computer vision classification tasks but now spans natural language processing, reinforcement learning, and multimodal applications. The k-shot framework enables researchers to systematically study how model performance scales with increasing data availability and to develop algorithms optimized for data-scarce scenarios. Common variations include 1-shot learning (single example per class), 5-shot learning (five examples per class), and k-shot generalization studies. For AI agents, k-shot capabilities determine how quickly systems can adapt to new domains, personalize to user preferences, and handle novel scenarios with minimal supervision.