• Practice
  • About 15 min
  • 5 practice questions

Artificial intelligence and machine learning: key terms, learning types and exam traps

Revise how AI, machine learning, deep learning and generative AI nest, match tasks to learning types, and spot overfitting, regression and inference in exam questions.

If something looks wrong, please tell us.

AI questions in competitive exams test vocabulary and classification: is this task supervised or unsupervised, classification or regression? What is the relationship between AI and deep learning? This lesson gathers the terms in one place and shows how to answer quickly.

How the ideas nest

Artificial intelligence is the widest field. Machine learning is the part of AI in which systems learn from data. Deep learning is the part of machine learning that uses neural networks with many layers. Generative AI - systems that create text, images, audio or code - is mostly built with deep learning. So every deep learning system is machine learning, but not every machine learning system is deep learning: a simple decision tree learns from data without any neural network.

Learning types: revision table

Three learning types
TypeDataTypical tasksMethods to recognise
SupervisedLabelledClassification (category), regression (number)Linear regression, decision trees, support vector machines
UnsupervisedUnlabelledClustering, finding unusual cases, reducing many features to a fewk-means clustering, principal component analysis
ReinforcementRewards and penalties from an environmentGames, robot control, sequences of decisionsAn agent learns a policy that earns the most reward

Key terms

Vocabulary to know
TermMeaning
Training dataThe examples a model learns from
ModelWhat learning produces: the learnt pattern used to make predictions
Parameters (weights)The numbers inside a model adjusted during training
InferenceUsing a trained model to produce an output for new input
OverfittingThe model memorises the training data, including its noise, and does poorly on new data
UnderfittingThe model is too simple to capture the pattern even in the training data
PromptThe instruction or question given to a generative model
HallucinationA confident but false or invented output from a generative model

Related application areas: computer vision (understanding images and video), natural language processing (understanding and producing human language), speech recognition and recommender systems (suggesting products or videos).

Common traps

  • Machine learning is not always deep learning. Deep learning is a subset.
  • Reinforcement learning is not supervised: there is no labelled correct answer for each step, only rewards.
  • Predicting a number is regression, even in supervised learning; predicting a category is classification.
  • Generative AI is not general AI. A model that writes fluent essays is still narrow AI.
  • High accuracy on training data proves little; judge a model on data it has never seen.

Worked example 1: label the tasks

  • Predict tomorrow's electricity demand in megawatts from past records → supervised, regression (a number).
  • Decide whether an X-ray image shows a fracture, using images marked by doctors → supervised, classification.
  • Find natural groups among one lakh unlabelled news articles → unsupervised, clustering.
  • Teach a program to manage traffic lights by rewarding shorter waiting times → reinforcement.

Worked example 2: spotting overfitting

A model scores 99% on its training data but only 70% on new test data. The gap of 29 percentage points shows it has memorised the training examples instead of learning the general pattern: overfitting. Remedies include more and more varied training data, a simpler model and stopping training earlier. If it had scored about 60% on both, the problem would be underfitting: the model is too simple.

AI ⊃ machine learning ⊃ deep learning. Category = classification; number = regression. Great on training, poor on new data = overfitting. Inference = using the trained model.

Key takeaways: place each term in the nesting of AI, ML and deep learning; match tasks to learning types by asking 'are there labels?' and 'is the output a category or a number?'. Judge models on unseen data.

Check your understanding

  1. 1. A model predicts the selling price of a house in rupees from its area and location, using past sales. This is:

    1. Supervised learning - classification
    2. Unsupervised learning - clustering
    3. Reinforcement learning
    4. Supervised learning - regression
    Show answer

    Answer: D. Past sales supply labels (the actual prices), so it is supervised, and the output is a number, so it is regression. Classification would predict a category.

  2. 2. A model scores 98% on its training data but 65% on new data. This most likely shows:

    1. Overfitting
    2. Underfitting
    3. Perfect generalisation
    4. Reinforcement
    Show answer

    Answer: A. A large gap between training and test performance means the model memorised the training data rather than learning the general pattern. That is overfitting.

  3. 3. Consider the statements: (1) All machine learning is deep learning. (2) Deep learning is a subset of machine learning. Which is/are correct?

    1. 1 only
    2. 2 only
    3. Both 1 and 2
    4. Neither 1 nor 2
    Show answer

    Answer: B. Deep learning, using many-layered neural networks, is one part of machine learning. Many machine learning methods, such as decision trees, are not deep learning, so statement 1 is wrong.

  4. 4. k-means clustering, which divides unlabelled data into groups, is an example of:

    1. Supervised learning
    2. Reinforcement learning
    3. Unsupervised learning
    4. Rule-based expert systems
    Show answer

    Answer: C. k-means works on data without labels and finds groups by itself, which makes it unsupervised learning.

  5. 5. In machine learning, 'inference' means:

    1. Collecting the training data
    2. Using a trained model to produce an output for new input
    3. Deleting a model after training
    4. Labelling the data by hand
    Show answer

    Answer: B. Training builds the model; inference is the stage at which the trained model is used on new input, such as answering a prompt or classifying a new photo.

Privacy · Terms · Refunds · Delivery