Common ML Terms

Machine learning introduces many new terms, and it is normal not to remember every definition the first time you encounter it. This glossary provides quick explanations of common machine learning and artificial intelligence concepts.

Accuracy

Accuracy measures the proportion of predictions a model gets correct. It can be useful for balanced classification problems, but may be misleading when some categories are much more common than others.

Algorithm

An algorithm is a procedure or method used to process data and solve a problem. In machine learning, an algorithm defines how a model learns patterns from data.

Artificial Intelligence

Artificial intelligence is the broader field of creating software systems that perform tasks associated with human reasoning, recognition, prediction, decision-making, or language use.

Batch

A batch is a group of training examples processed together during one training step. Batch processing can make training more efficient and help manage memory use.

Classification

Classification is a machine learning task in which a model predicts a category or label, such as whether a message belongs to one category or another.

Data Drift

Data drift occurs when the characteristics or distribution of incoming data changes compared with the data used during training. This change may reduce model performance and should be investigated.

Dataset

A dataset is a collection of information used to train, validate, test, or analyze a machine learning model. Datasets may contain text, images, measurements, records, audio, or other data.

Deep Learning

Deep learning is a branch of machine learning that uses neural networks with multiple layers to learn complex patterns. It is used for tasks involving images, audio, language, video, and other types of data.

Deployment

Deployment is the process of making a trained model available for use in an application, service, device, website, programming interface, or automated workflow.

Embeddings

Embeddings are numerical representations of data that capture useful relationships or similarities. They are commonly used for search, recommendations, clustering, and language applications.

Epoch

An epoch is one complete pass through the training dataset. A model usually processes the data over multiple epochs while its parameters are adjusted.

Feature

A feature is an individual piece of information used by a model to make a prediction. Examples include age, temperature, word frequency, image values, or the number of previous events.

Feature Engineering

Feature engineering is the process of creating, selecting, transforming, or combining inputs so they are more useful for a machine learning model.

Fine-Tuning

Fine-tuning is the process of adapting a pretrained model to a specific task by training it further with task-specific data.

Generalization

Generalization is a model’s ability to perform well on new data rather than only on the examples used during training.

Hyperparameters

Hyperparameters are settings that control how a model or training process behaves. Examples include learning rate, batch size, tree depth, model complexity, and number of epochs.

Inference

Inference is the process of using a trained model to produce predictions or outputs for new data. Training teaches the model, while inference is when the model performs its intended task.

Large Language Model

A large language model, or LLM, is a neural network trained on large collections of language data. It can generate, summarize, classify, transform, and analyze text, although its outputs still require appropriate evaluation.

Labels

Labels are the known categories, values, or outcomes associated with examples in a supervised learning dataset. They provide the expected answers used during training.

Loss Function

A loss function measures the difference between a model’s prediction and the desired result during training. Optimization methods adjust the model to reduce this difference.

Machine Learning

Machine learning is an approach to software development in which models learn patterns from data and use those patterns to make predictions, classifications, decisions, or other outputs.

Machine Learning Stack

A machine learning stack is the collection of data systems, features, models, tools, workflows, infrastructure, deployment processes, and monitoring used to build and operate machine learning applications.

Model

A model is the learned representation produced by a machine learning process. It uses patterns found in data to produce predictions or other outputs for new inputs.

Model Drift

Model drift occurs when a model’s performance declines because the relationship between its inputs and the desired outcomes has changed over time.

Monitoring

Monitoring is the process of observing a deployed model and its surrounding application. It may track prediction quality, data changes, errors, response times, resource use, and other signals.

Neural Network

A neural network is a machine learning model made up of connected layers that transform data and learn patterns. Neural networks are the foundation of deep learning.

Overfitting

Overfitting occurs when a model memorizes details of its training data instead of learning patterns that generalize. It performs well on training examples but poorly on new data.

Parameters

Parameters are internal values adjusted during training. They represent patterns learned from the data and influence the model’s predictions.

Precision

Precision measures how many of the examples predicted as positive are actually positive. It is useful when false positive predictions are especially costly.

Recall

Recall measures how many of the actual positive examples a model correctly identifies. It is useful when missing positive cases is especially costly.

Regression

Regression is a machine learning task in which a model predicts a continuous numerical value, such as a price, temperature, quantity, or measurement.

Supervised Learning

Supervised learning trains a model using examples that include known labels or outcomes. It is commonly used for classification and regression.

Test Set

A test set is a portion of data reserved for final evaluation. It should not be used directly to train the model or repeatedly choose its settings.

Training

Training is the process of providing data to a machine learning algorithm so it can adjust internal parameters and learn useful patterns.

Underfitting

Underfitting occurs when a model is too simple, insufficiently trained, or provided with weak features. It performs poorly on both training data and new examples.

Unsupervised Learning

Unsupervised learning uses data without predefined labels. The model searches for groups, relationships, unusual observations, or other patterns within the data.

Validation Set

A validation set is a portion of data used during development to compare models, adjust settings, and guide improvements before final testing.