Supervised learning
Learns relationships from examples with known target labels or values.
Related: Unsupervised learning · Reinforcement learning
AI-generated content. Questions and study notes were written and checked by AI agents from public documentation. They may contain errors and are not real exam questions. Verify with the linked sources. Learn more
Key terms, each linked to its source.
42 results · content in English
Learns relationships from examples with known target labels or values.
Related: Unsupervised learning · Reinforcement learning
Finds structure or groups in data without target labels.
Related: Supervised learning · Reinforcement learning
An agent learns actions through rewards to improve cumulative outcomes.
Related: Supervised learning · Unsupervised learning
Predicts a continuous numeric value, such as a price or duration.
Related: Classification · Clustering
Groups similar records when the desired group labels are not supplied.
Related: Regression · Classification
Transforms raw data into useful model inputs, such as deriving tenure from a signup date.
Related: Hyperparameter tuning
Compares training configurations using a validation metric; hyperparameters are chosen outside the learned model weights.
Related: Feature engineering
The proportion of all predictions that are correct; class imbalance can make it misleading.
Related: F1 score · Mean absolute error · Root mean squared error
The harmonic mean of precision and recall, balancing false-positive and false-negative concerns.
Related: Accuracy · Mean absolute error · Root mean squared error
Average absolute distance between predicted and actual numeric values, in the target's units.
Related: Accuracy · F1 score · Root mean squared error
Square root of average squared prediction error; large errors receive greater penalty than with MAE.
Related: Accuracy · F1 score · Mean absolute error
Generated content that is unsupported or false despite sounding plausible.
Related: Knowledge cutoff · Nondeterminism
A limit on information available from a model's training data; later facts require an external source or updating.
Related: Hallucination · Nondeterminism
The same input can produce different outputs across runs.
Related: Hallucination · Knowledge cutoff
Nucleus sampling restricts token choices to a set reaching a cumulative probability threshold.
Related: Temperature · Top K
Asks a model to perform a task without giving task examples in the prompt.
Related: Prompt chaining
Breaks work into a sequence of prompts, passing intermediate results to later steps.
Related: Zero-shot prompting
Updates a pretrained model using task-specific training examples to adapt its behavior.
Related: Continued pre-training · Distillation · Retrieval Augmented Generation
Further trains a pretrained model on domain text to adapt its knowledge and language patterns.
Related: Fine-tuning · Distillation · Retrieval Augmented Generation
Trains a smaller student model using a larger teacher's outputs or behavior.
Related: Fine-tuning · Continued pre-training · Retrieval Augmented Generation
Retrieves relevant external information and includes it in model context at inference time; retrieval does not update model weights.
Related: Fine-tuning · Continued pre-training · Distillation
A family of reference-overlap metrics commonly used to evaluate summaries.
Related: BLEU · BERTScore · LLM-as-a-judge
A reference-based n-gram precision metric commonly associated with machine translation.
Related: ROUGE · BERTScore · LLM-as-a-judge
Compares contextual token embeddings to estimate semantic similarity to reference text.
Related: ROUGE · BLEU · LLM-as-a-judge
Whether an answer is supported by its retrieved context, distinct from whether the context itself is factually correct.
Related: Context relevance
Helps people understand why an AI system produced a behavior or result.
Related: Interpretability · Transparency · Local explanation
The degree to which a model's internal workings can be understood directly.
Related: Explainability · Transparency · Local explanation
Communicates AI use, intended purposes, limitations, and other information stakeholders need to make informed choices.
Related: Explainability · Interpretability · Local explanation
Explains one prediction or output, rather than aggregate behavior across the model.
Related: Explainability · Interpretability · Transparency
Replaces identifiers with aliases while retaining a protected way to re-identify records.
Related: Anonymization · Differential privacy
Irreversibly removes or generalizes identifiers so records no longer relate to an identifiable person.
Related: Pseudonymization · Differential privacy
Uses calibrated randomness with a privacy guarantee to limit what analysis reveals about any individual.
Related: Pseudonymization · Anonymization