Machine Learning Vocabulary in English
20 essential machine learning terms with meanings, example sentences, and free interactive games — ideal for B2–C1 learners working in or studying technology.
What You'll Learn
- ✅ Core ML terms used in technical documentation, papers, and job descriptions
- ✅ The difference between supervised and unsupervised learning explained in plain English
- ✅ How to discuss model performance using words like accuracy, precision, and recall
- ✅ Key pipeline and deployment vocabulary for professional technical communication
Pedagogically reviewed by LexFizz Team
Machine learning has become one of the most discussed fields in modern technology, and its specialist vocabulary is increasingly important for anyone working in or around software, data, and artificial intelligence. Whether you are reading research papers, writing technical documentation, attending meetings about AI projects, or simply trying to understand the news, knowing these core machine learning terms, as defined by the Oxford Learner's Dictionaries, will help you communicate confidently in English.
The vocabulary of machine learning draws on mathematics, statistics, computer science, and engineering. Many terms — such as gradient descent, backpropagation, and loss function — describe the internal processes by which a model learns from data. Others — such as deployment, pipeline, and benchmark — describe the practical engineering work of taking a model from research into production.
At B2 level, focus on understanding the meaning of each term in context. You do not need to understand the underlying mathematics to use these words correctly in professional conversation. At C1 level, practise using these terms when discussing trade-offs: for example, explaining why improving recall sometimes reduces precision, or why regularisation is used to prevent overfitting.
Machine learning English is also full of collocations — fixed word combinations — that are worth learning as units: train a model, evaluate performance, tune hyperparameters, deploy to production. Memorising these combinations alongside the individual words will make your technical English sound natural and professional.
Word List
| Word | Meaning | Example Sentence |
|---|---|---|
| supervised | a type of ML where the model learns from labelled training examples | Supervised learning requires a large set of correctly labelled data. |
| unsupervised | ML where the model finds patterns in data without labels | Unsupervised learning can reveal hidden clusters in customer data. |
| training data | the dataset used to teach a machine learning model | The quality of training data has a huge impact on model accuracy. |
| label | the correct answer or category assigned to a training example | Each image in the dataset was given a label identifying the object shown. |
| epoch | one complete pass through the entire training dataset | The model was trained for fifty epochs before performance plateaued. |
| gradient descent | an algorithm that minimises error by adjusting model weights step by step | Gradient descent is the foundation of most modern neural network training. |
| loss function | a measure of how far a model's predictions are from the correct answers | A lower loss function value means the model is making fewer errors. |
| hyperparameter | a configuration setting chosen before training that controls how the model learns | The learning rate is one of the most important hyperparameters to tune. |
| cross-validation | a technique for evaluating a model by testing it on multiple subsets of data | Cross-validation helps ensure the model performs well on unseen data. |
| regularisation | a method of reducing overfitting by penalising model complexity | L2 regularisation was applied to prevent the model from memorising the training data. |
| backpropagation | the algorithm used to calculate how to update weights during neural network training | Backpropagation computes the gradient of the loss with respect to each weight. |
| tensor | a multi-dimensional array used to represent data in ML frameworks | The input image was converted into a tensor before being fed to the model. |
| embedding | a numerical representation of a word, image, or object in vector space | Word embeddings allow the model to capture semantic relationships between terms. |
| accuracy | the proportion of correct predictions made by a model | The model achieved 94% accuracy on the test set. |
| precision | the proportion of positive predictions that are actually correct | High precision means few false positives in the model's output. |
| recall | the proportion of actual positives that the model correctly identifies | In medical diagnosis, high recall is critical to avoid missing real cases. |
| benchmark | a standard test used to compare the performance of different models | The new architecture outperformed all previous models on the ImageNet benchmark. |
| inference | the process of using a trained model to make predictions on new data | Inference speed is crucial for real-time applications like voice assistants. |
| deployment | the process of making a trained model available for use in a real-world system | The team spent two weeks on deployment before the model went live. |
| pipeline | a sequence of data processing steps from raw input to final prediction | The data pipeline handles cleaning, transformation, and feature extraction automatically. |
Practice with These Exercises
Wordsearch
Find ML terms hidden in the grid
Anagram
Unscramble the letters to spell each term
Hangman
Guess the ML word letter by letter
Flash Cards
Review ML vocabulary with spaced repetition
Practice What You've Learned
LexFizz has 30 free interactive exercises — no sign-up needed.
Browse All Exercises →Related Vocabulary Topics
Frequently Asked Questions
What is the difference between supervised and unsupervised learning?
In supervised learning, the model is trained on labelled data — each input has a correct answer attached to it, like an image labelled "cat" or "dog". The model learns to map inputs to those correct outputs. In unsupervised learning, the data has no labels and the model must discover structure on its own, such as grouping customers into segments or detecting anomalies in network traffic. A third category, reinforcement learning, trains an agent through rewards and penalties rather than labelled examples.
What is the difference between precision and recall?
Precision measures how many of the items the model flagged as positive are actually positive — it answers "when the model says yes, how often is it right?" Recall measures how many of the actual positives the model successfully found — it answers "of all the real positives, how many did the model catch?" There is usually a trade-off between the two: increasing recall often reduces precision, and vice versa. In spam filtering, high precision avoids putting real emails in spam; in medical screening, high recall avoids missing true cases of disease.
Why is training data so important in machine learning?
Training data is the foundation on which a model learns. A model can only learn patterns that exist in its training data, so if the data is biased, incomplete, or noisy, the model will make poor predictions. The phrase "garbage in, garbage out" is commonly used in data science to emphasise this dependency. Collecting, cleaning, and labelling high-quality training data is often the most time-consuming and expensive part of an ML project.
What does epoch mean in machine learning?
An epoch is one complete pass through the entire training dataset. During each epoch, the model sees every training example once, updates its weights based on the errors it makes, and gradually improves. Training typically requires many epochs — sometimes dozens or hundreds — before the model converges on a good solution. Monitoring the loss function across epochs helps you decide when to stop training.
What is a hyperparameter and how is it different from a parameter?
Parameters are the internal values that a model learns during training, such as the weights in a neural network. Hyperparameters are the external settings you choose before training begins, such as the learning rate, the number of layers, or the batch size. Hyperparameter tuning — systematically testing different combinations — is an important step in getting the best performance from a model.
What is regularisation and why is it used?
Regularisation is a set of techniques used to prevent overfitting — the problem where a model learns the training data too well and fails to generalise to new examples. Common methods include L1 regularisation (which can set some weights to zero), L2 regularisation (which penalises large weights), and dropout (which randomly turns off neurons during training). The result is a simpler model that performs better on unseen data.
What is the difference between inference and training?
Training is the process of teaching a model by showing it many examples and adjusting its weights until it makes accurate predictions. Inference is the process of using the already-trained model to make predictions on new, real-world data. Training is computationally expensive and done in advance; inference must be fast enough for real-time use and is what actually runs in production systems.
What is an embedding in NLP or machine learning?
An embedding is a way of representing something — a word, a sentence, an image, or a user — as a list of numbers (a vector) in a multi-dimensional space. Words with similar meanings end up close together in this space. For example, "king" and "queen" will have similar embeddings, and "cat" and "dog" will be near each other. Embeddings allow machine learning models to work with text and other non-numeric data.
What is a machine learning pipeline?
A machine learning pipeline is a sequence of automated steps that takes raw data and produces a final prediction or output. A typical pipeline includes: data ingestion, data cleaning, feature engineering, model training, evaluation, and deployment. Pipelines make the ML workflow reproducible and easier to maintain — if the data or model changes, you can rerun the whole sequence consistently.
How is machine learning vocabulary used in job interviews?
Technical interviews for ML roles often ask candidates to explain concepts like backpropagation, cross-validation, or the bias-variance trade-off in plain English. Being able to define these terms clearly, give an example, and explain why they matter shows both technical knowledge and communication skill. Practising definitions out loud — not just reading them — is the best preparation for this type of interview.