Concept dictionary
Training Data
Training data is the collection of examples used during training to adjust a model's parameters and shape its behavior on a defined objective.
- Review state
- Reviewed in POC
- Freshness
- Current
- Version
- 2
Also called
- Training Data
Remember it this wayTraining data supplies examples from which the model learns patterns.
In plain language
The examples a model learns from influence which patterns it can recognize and reproduce. Data quality, coverage, labeling, provenance, and representativeness all matter alongside the training method.
Why it matters
Training data can introduce useful knowledge as well as gaps, bias, legal risk, and outdated assumptions. It is one major contributor to behavior, not the only one.
Example
A document classifier is trained on earlier messages labeled by category so it can classify new messages.
Common misconception
Training data is not the single explanation for every output. Architecture, objectives, fine-tuning, prompts, retrieval, tools, and deployment controls also influence behavior.
Related terms
Sources
- NIST AI 100-2 E2025, Adversarial Machine LearningNIST terminology for training data and threats affecting it.
- NIST AI 100-1, Artificial Intelligence Risk Management Frameworkgovernance and measurement guidance for data used across the AI lifecycle.
Learn more
- Trace origins with [[data-provenance|Data Provenance]] and downstream effects with [[bias|Bias]].