Data & Training
15 Terms
A–Z Linked
Data & Training
Concepts related to datasets, model training, evaluation, and data preparation.
Cross-Validation
📘 Definition
Cross-validation is an evaluation technique that repeatedly divides data into different training and validation sets to estimate model performance more reliably.
🪄 Simple Explanation
Cross-validation provides a better estimate of model quality.
🔑 Why It Matters
It reduces the risk of relying on a single data split.
💡 Example
Researchers compare models using five-fold cross-validation.
See Also
Data Annotation
📘 Definition
Data annotation is the process of adding labels or descriptions to data so it can be used for supervised learning.
🪄 Simple Explanation
Annotation gives AI the correct answers during training.
🔑 Why It Matters
High-quality annotation improves supervised models.
💡 Example
Images are labelled with the objects they contain.
Data Augmentation
📘 Definition
Data augmentation creates additional training examples by making controlled changes to existing data.
🪄 Simple Explanation
It expands datasets without collecting new data.
🔑 Why It Matters
Augmentation often improves model robustness.
💡 Example
Images are rotated and flipped during training.
See Also
Data Cleaning
📘 Definition
Data cleaning is the process of correcting, removing, or standardising inaccurate, incomplete, or inconsistent data before analysis or model training.
🪄 Simple Explanation
Data cleaning improves data quality.
🔑 Why It Matters
Cleaner data usually leads to better AI performance.
💡 Example
Duplicate customer records are merged before training a model.
See Also
Data Drift
📘 Definition
Data drift occurs when the characteristics of incoming data change over time, reducing model performance.
🪄 Simple Explanation
AI may become less accurate as data changes.
🔑 Why It Matters
Monitoring drift keeps models reliable.
💡 Example
Shopping behaviour changes after a major holiday season.
See Also
Data Pipeline
📘 Definition
A data pipeline is a sequence of processes that collects, prepares, transforms, and delivers data for AI systems.
🪄 Simple Explanation
Pipelines move data from source to model.
🔑 Why It Matters
Well-designed pipelines improve efficiency.
💡 Example
Sensor readings are cleaned before reaching a prediction model.
See Also
Data Quality
📘 Definition
Data quality describes how accurate, complete, consistent, and reliable a dataset is for its intended purpose.
🪄 Simple Explanation
High-quality data leads to better AI.
🔑 Why It Matters
Poor-quality data often produces poor results.
💡 Example
A dataset with missing values is improved before training.
See Also
Labelled Data
📘 Definition
Labelled data consists of examples that include both input data and the correct output used for supervised learning.
🪄 Simple Explanation
Labelled data teaches AI the correct answers.
🔑 Why It Matters
It is essential for supervised learning.
💡 Example
Images labelled as cats and dogs train an image classifier.
See Also
Model Monitoring
📘 Definition
Model monitoring is the continuous observation of an AI model after deployment to ensure that it continues to perform accurately and reliably.
🪄 Simple Explanation
Monitoring checks whether an AI system is still working well.
🔑 Why It Matters
Ongoing monitoring helps detect performance problems early.
💡 Example
A bank monitors its fraud detection model to ensure prediction quality remains high.
See Also
Preprocessing
📘 Definition
Preprocessing prepares raw data for AI by cleaning, transforming, and organising it before training begins.
🪄 Simple Explanation
Preprocessing makes data suitable for learning.
🔑 Why It Matters
Well-prepared data improves model performance.
💡 Example
Dates are converted into a standard format before training.
See Also
Sampling
📘 Definition
Sampling is the process of selecting a representative subset of data from a larger dataset.
🪄 Simple Explanation
Sampling allows AI to learn from manageable amounts of data.
🔑 Why It Matters
Representative samples improve model reliability.
💡 Example
A survey uses a random sample rather than the entire population.
See Also
Testing
📘 Definition
Testing evaluates a completed model using unseen data to estimate real-world performance.
🪄 Simple Explanation
Testing checks whether AI generalises well.
🔑 Why It Matters
Reliable testing prevents unrealistic expectations.
💡 Example
Researchers report final accuracy on a separate test set.
See Also
Training
📘 Definition
Training is the process of teaching an AI model by exposing it to data and adjusting its parameters.
🪄 Simple Explanation
Training is how AI learns.
🔑 Why It Matters
Training determines what a model knows.
💡 Example
A language model trains on billions of words.
See Also
Unlabelled Data
📘 Definition
Unlabelled data contains inputs without known answers or categories.
🪄 Simple Explanation
Unlabelled data supports unsupervised learning.
🔑 Why It Matters
Large collections of raw data are often unlabelled.
💡 Example
Customer browsing records are analysed without predefined groups.
See Also
Validation
📘 Definition
Validation measures model performance during development so improvements can be made before final testing.
🪄 Simple Explanation
Validation guides model improvement.
🔑 Why It Matters
It helps choose the best model.
💡 Example
Developers compare several models using validation data.