Programmer's Picnic
AI-ML Terminology Lesson
Start Video Core Terms ML Terms Deep Learning Quick Revision
📘 Ground-level lesson

AI-ML Terminology from Ground Level

This page explains the words you keep hearing in AI and Machine Learning, starting from absolute beginner level. The goal is not just to define terms, but to make them easy to understand and remember.

Beginner-friendly Comprehensive Examples included Plain English
Jump to Topics Revision Sheet

📺 A man crossing a road| AI-ML

How do you cross a street?
Find out the speed of every vehicle in the area. Then measure the width of the road blah blah blah..

No, you do it all by experience and common sense. this is learning as opposed to calculating

Open in YouTube

📺 Discussing AI-ML

Introducing AI-ML terms

AI-ML Terminology Intro Video

Open in YouTube

How to read this lesson

Read the early sections carefully. Later terms become much easier once you clearly understand words like data, model, training, feature, label, and prediction.

In AI-ML, many difficult-looking words are just simple ideas with technical names.
  • First understand the basic idea.
  • Then read the example.
  • Then read the analogy.
  • Then move to the next term.

Part 1: Ground-level ideas

These are the most important starting words.

FOUNDATION

1. Artificial Intelligence (AI)

Artificial Intelligence means making a computer system perform tasks that usually need human intelligence. These tasks may include understanding language, recognizing images, making decisions, solving problems, or learning from experience.

Example: A chatbot answering questions, a phone unlocking by face, or a map suggesting the best route.
Analogy: AI is like a broad umbrella. Many smaller topics come under it, and Machine Learning is one of them.
FOUNDATION

2. Machine Learning (ML)

Machine Learning is a part of AI where a computer learns patterns from data instead of being told every rule manually.

In old-style programming, we write rules directly. In Machine Learning, we give examples, and the system learns the rule by itself.

Example: Instead of writing hundreds of spam rules by hand, you show many spam and non-spam messages to the model and let it learn the difference.
Analogy: It is like teaching a child with examples instead of writing a giant rulebook.
VERY IMPORTANT

3. Data

Data is the raw information used by the system. It can be numbers, text, images, sound, clicks, marks, prices, or anything that can be stored and processed.

Example: Student marks, attendance, house prices, product reviews, and email messages are all data.
Analogy: If AI is a car, data is the fuel.
VERY IMPORTANT

4. Dataset

A dataset is a collection of related data, usually arranged in a table or structured format.

Example: A file containing columns like Name, Study Hours, Attendance, and Marks is a dataset.
Name Hours Attendance Marks Aman 2 80 45 Riya 5 90 70 Karan 6 95 82
CORE TERM

5. Feature

A feature is an input piece of information used by the model to make a decision or prediction.

Example: In a marks predictor, study hours and attendance can be features.
Analogy: Features are clues given to the model.
CORE TERM

6. Label / Target

A label, also called target, is the correct answer the model is trying to learn.

Example: If the input is study hours and attendance, then the output marks may be the label.
Analogy: The label is the answer at the back of the book.
CORE TERM

7. Model

A model is the learned system that takes inputs and gives outputs. It is the thing that has learned from data.

Example: After training on many examples of study hours and marks, the model can guess marks for a new student.
Analogy: The model is like the trained brain of the system.
CORE TERM

8. Training

Training is the process of showing data to the model so it can learn patterns.

Example: You give the model many examples of hours studied and actual marks, and it learns the relationship.
Analogy: Training is like practice before an exam.
CORE TERM

9. Testing

Testing means checking whether the trained model works well on new data it did not see during training.

Example: You hide some records from the model, then later use those hidden records to check performance.
Analogy: Training is practice; testing is the real exam.
CORE TERM

10. Prediction

Prediction is the output the model gives for new input data.

Example: “Predicted marks = 78” or “Predicted class = spam”.

Part 2: Core Machine Learning terminology

ML TASK

11. Regression

Regression is used when the output is a number.

Example: Predicting marks, price, temperature, salary, or sales.
Analogy: Regression answers “how much?”
ML TASK

12. Classification

Classification is used when the output is a category or class.

Example: Spam / not spam, pass / fail, cat / dog, positive / negative.
Analogy: Classification answers “which group?”
LEARNING TYPE

13. Supervised Learning

Supervised learning means training with inputs and correct outputs together.

Example: A table with study hours as input and marks as correct output.
Analogy: A teacher is supervising because the correct answers are known.
LEARNING TYPE

14. Unsupervised Learning

Unsupervised learning means the system only gets data, not correct answers, and it tries to discover structure or patterns by itself.

Example: Grouping customers by similar buying patterns.
Analogy: It is like sorting similar objects without being told the category names.
METHOD

15. Algorithm

An algorithm is the method or technique used to train the model.

Example: Linear Regression, Decision Tree, KNN, Naive Bayes.
Analogy: The algorithm is the recipe. The model is the final dish prepared using that recipe.
COMMON PROBLEM

16. Overfitting

Overfitting happens when the model memorizes training data too much and fails on new data.

Example: The model gets almost every training example right but performs poorly on testing data.
Analogy: A student memorized solved questions only, but cannot handle a fresh question in the exam.
COMMON PROBLEM

17. Underfitting

Underfitting happens when the model is too weak to learn the pattern properly.

Example: It performs badly on both training data and testing data.
Analogy: A student did not study enough even for practice questions.
EVALUATION

18. Accuracy

Accuracy tells us how many predictions were correct out of the total.

Accuracy = Correct Predictions / Total Predictions
Example: If 90 out of 100 predictions are correct, the accuracy is 90%.
EVALUATION

19. Loss

Loss measures how wrong the model’s predictions are. Smaller loss is better.

Example: If the actual marks are 80 but the model predicts 20, the loss is high.
Analogy: Loss is like error pain. The training process tries to reduce that pain.
TRAINING TERM

20. Epoch

An epoch means one full pass through the training data during learning.

Example: If your dataset has 100 rows and the model reads all 100 once, that is one epoch.
ADVANCED IDEA

21. Bias and Variance

Bias is error from being too simple. Variance is error from being too sensitive to the training data.

Simple summary: High bias often causes underfitting. High variance often causes overfitting.

Part 3: Deep Learning terminology

DEEP LEARNING

22. Deep Learning

Deep Learning is a part of Machine Learning that uses layered neural networks to learn complex patterns.

Example: Image recognition, speech recognition, advanced language models.
Analogy: If basic ML is like solving with a few simple steps, deep learning is like solving through many layers of thinking.
DEEP LEARNING

23. Neural Network

A neural network is a model made of connected layers that transform input step by step into output.

Example: It can take pixel values from an image and finally predict “cat”.
NETWORK PART

24. Input Layer

The input layer is where the model receives data.

NETWORK PART

25. Hidden Layer

Hidden layers are the internal processing layers where the network learns patterns.

NETWORK PART

26. Output Layer

The output layer gives the final prediction.

NETWORK TERM

27. Weights

Weights are numbers inside the model that tell it how important different inputs are.

NETWORK TERM

28. Activation Function

An activation function helps the neural network decide how strongly a neuron should respond.

Common names: ReLU, Sigmoid, Tanh.
TRAINING TERM

29. Backpropagation

Backpropagation is the method used to send the error backward through the network and adjust weights.

TRAINING TERM

30. Optimizer

An optimizer is the method used to update the model so that loss becomes smaller.

Example: Gradient Descent, Adam.

Important distinction

  • AI is the broad field.
  • ML is one part of AI.
  • Deep Learning is one part of ML.

Three questions to ask

  • What is the input?
  • What is the output?
  • What pattern must the model learn?

How terms connect

  • Dataset contains data.
  • Features are inputs.
  • Labels are answers.
  • Training builds the model.
  • Testing checks the model.

Quick Revision Sheet

Fast definitions

  • AI: making computers do intelligent tasks
  • ML: learning patterns from data
  • Data: raw information
  • Dataset: collection of data
  • Feature: input clue
  • Label: correct answer
  • Model: learned system
  • Training: learning from examples
  • Testing: checking on new examples
  • Prediction: model output

Fast task types

  • Regression: predict a number
  • Classification: predict a category
  • Supervised: train with answers
  • Unsupervised: find patterns without answers
  • Overfitting: memorized too much
  • Underfitting: learned too little
  • Accuracy: how many were correct
  • Loss: how wrong the model is
  • Deep Learning: ML with layered neural networks
Best learning rule: whenever you hear a new term, ask, “Is this an input, an output, a method, a model, a problem, or a result?”

Final message

AI-ML terminology looks heavy only at the beginning. Once you connect each word to a simple idea, the subject becomes much easier. Learn the words slowly, connect them to examples, and use them while building small projects.

Real understanding grows when you read the term, see it in code, and use it in a project.