Beginner’s Course Material
Artificial Intelligence From Zero
A comprehensive, structured study guide covering everything a beginner needs to understand, learn, and apply Artificial Intelligence โ from foundational concepts to real-world applications.
10 ChaptersStudy Units
60+ TopicsIn-Depth Coverage
BeginnerDifficulty Level
Self-PacedLearning Format
01
Chapter One ยท Foundation
Introduction to Artificial Intelligence
Before writing a single line of code, you must understand what Artificial Intelligence actually is โ its definition, history, types, and how it fits into our world today.
Artificial Intelligence is the science of making machines capable of performing tasks that would normally require human intelligence โ reasoning, learning, problem-solving, perception, and language understanding.
โ John McCarthy, father of AI, 1956
Topics in this Chapter
Definition of AI, differences between AI and traditional software, and how AI systems make decisions. Understanding the concept of intelligence in machines vs humans.
From the Turing Test (1950) to expert systems in the 1980s, the AI winters, the machine learning revolution, and the rise of deep learning and large language models. Key milestones you need to know.
Narrow AI (ANI) does one task well. General AI (AGI) can reason across domains like a human. Super AI (ASI) surpasses human intelligence. Understanding where we are today and the practical implications of each.
The often-confused hierarchy: AI is the broadest concept, Machine Learning is a subset of AI, and Deep Learning is a subset of ML. Learn the precise distinctions with real-world examples.
The fundamental idea that machines learn from data rather than being explicitly programmed. Introduction to training, patterns, and generalisation โ the core engine behind all modern AI.
Spam filters, recommendation engines, facial recognition, voice assistants, self-driving cars, medical diagnosis, and language models. Grounding abstract concepts in familiar technology.
Overview of major companies (OpenAI, Google DeepMind, Anthropic, Meta AI), academic institutions, and open-source communities that shape the field. Understanding the AI ecosystem.
1.8 Setting Up Your Learning Environment
Installing Python, setting up Jupyter Notebooks or Google Colab, and creating your first AI-ready coding environment. Step-by-step walkthrough for non-programmers.
๐ Key Terms
Artificial Intelligence
Machine Learning
Deep Learning
Turing Test
Narrow AI
Training Data
Generalisation
02
Chapter Two ยท Foundation
Mathematics for Artificial Intelligence
You don’t need to be a mathematician โ but a solid grasp of the core math concepts will unlock your understanding of how AI algorithms actually work under the hood.
Topics in this Chapter
Scalars, vectors, matrices, and tensors โ the data structures AI operates on. Matrix multiplication, dot products, transpose, and inverse. Why neural networks are essentially giant matrix operations.
2.2 Calculus for Machine Learning
Derivatives and gradients โ how models measure and minimise their own errors. The chain rule, partial derivatives, and gradient descent explained intuitively with visual examples.
2.3 Probability & Statistics Fundamentals
Probability distributions, mean, median, variance, and standard deviation. Bayes’ theorem, conditional probability, and why probabilistic thinking is central to AI predictions and uncertainty.
2.4 Gradient Descent โ The Engine of Learning
How models improve iteratively by following the gradient downhill toward minimum error. Batch, stochastic, and mini-batch gradient descent. Learning rates and their critical impact on training.
2.5 Loss Functions & Cost Functions
How we measure how wrong a model is โ Mean Squared Error (MSE), Cross-Entropy Loss, and others. The difference between loss and cost functions, and why the choice matters enormously.
2.6 Vectors in High Dimensions
Understanding data points as vectors in multi-dimensional space. Distance measures (Euclidean, cosine similarity), what “similar data” means geometrically, and the curse of dimensionality.
2.7 Introduction to Information Theory
Entropy, information gain, and KL divergence โ concepts that explain how AI models measure uncertainty, make decisions under ambiguity, and are trained with cross-entropy loss.
๐ Math Tools to Use
You can use NumPy for linear algebra, SymPy for calculus, and SciPy for statistics. Don’t calculate by hand โ let libraries do the heavy lifting while you understand the concepts.
Python โ Basic Linear Algebra with NumPy
import numpy as np
# Create a matrix
A = np.array([[1, 2], [3, 4]])
# Matrix multiplication (dot product)
B = np.array([[5, 6], [7, 8]])
result = np.dot(A, B)
# Transpose a matrix
A_T = A.T
print("Matrix A transposed:", A_T)
03
Chapter Three ยท Foundation
Python Programming for AI
Python is the language of AI. Its simplicity, rich ecosystem, and community make it the universal tool for every AI practitioner โ from beginners to researchers at the world’s top labs.
Topics in this Chapter
3.1 Python Basics Refresher
Variables, data types, loops, conditionals, functions, and classes. Understanding Python syntax well enough to read and write AI code without confusion.
3.2 NumPy โ Numerical Computing
Arrays, array operations, broadcasting, slicing, and reshaping. NumPy is the backbone of all AI libraries โ mastering it unlocks everything that follows.
3.3 Pandas โ Data Manipulation
DataFrames, loading CSV files, filtering, grouping, handling missing values, and merging datasets. Real AI work is 80% data wrangling โ Pandas is your primary tool.
3.4 Matplotlib & Seaborn โ Data Visualisation
Plotting line charts, bar charts, scatter plots, heatmaps, and histograms. Visualising data before and after modelling is a professional habit that uncovers problems early.
3.5 Working with Data โ Loading, Cleaning, Exploring
Exploratory Data Analysis (EDA), detecting outliers, normalising features, encoding categorical variables, and train/test splitting. The data preparation pipeline every AI practitioner follows.
3.6 Jupyter Notebooks & Google Colab
How to use interactive notebook environments for AI experiments. Running cells, documenting your work, and using free GPU acceleration on Google Colab for model training.
3.7 Object-Oriented Programming for AI
Classes, objects, inheritance, and encapsulation โ understanding how AI libraries like PyTorch use OOP so you can extend and customise models effectively.
3.8 Working with APIs & Data Sources
Making HTTP requests, consuming REST APIs, working with JSON, and scraping data ethically. Most real-world AI projects source data from external APIs and the web.
04
Chapter Four ยท Core Concepts
Machine Learning Fundamentals
Machine Learning is how AI learns from data. Rather than being programmed with explicit rules, ML models discover patterns through examples โ and this chapter explains every key concept in that process.
Types of Machine Learning
๐ซ
Supervised Learning
Learn from labelled examples. Given inputs and correct outputs, the model learns the mapping. Used for classification and regression tasks.
Most Common
๐
Unsupervised Learning
Find hidden patterns in unlabelled data. Clustering, dimensionality reduction, and anomaly detection are key applications.
๐ฎ
Reinforcement Learning
An agent learns by taking actions in an environment and receiving rewards or penalties. Famously used in game-playing AI and robotics.
Topics in this Chapter
4.1 The ML Workflow โ End to End
Problem definition โ data collection โ data preprocessing โ feature engineering โ model selection โ training โ evaluation โ deployment. Understanding the complete lifecycle before touching a single algorithm.
4.2 Linear Regression
Predicting a continuous value (e.g. house price) from features. The simplest ML model โ understanding it deeply reveals the core mechanics of all models: weights, bias, loss, and optimisation.
4.3 Logistic Regression & Classification
Predicting a category (spam/not spam) using the sigmoid function. Decision boundaries, probability outputs, and binary vs multiclass classification.
4.4 Decision Trees & Random Forests
Tree-based models that split data by asking yes/no questions. Random Forests improve accuracy by combining hundreds of trees (ensemble learning) โ one of the most powerful beginner-friendly algorithms.
4.5 Support Vector Machines (SVM)
Finding the optimal boundary (hyperplane) that best separates classes with maximum margin. Kernel trick for non-linear separation and when to prefer SVMs over neural networks.
4.6 K-Nearest Neighbours (KNN)
Classify a data point based on the majority label of its K nearest neighbours. One of the most intuitive ML algorithms โ and a perfect illustration of how distance metrics work in practice.
4.7 K-Means Clustering
Unsupervised grouping of data into K clusters by iteratively assigning points to the nearest cluster centre. Applications in customer segmentation, document grouping, and image compression.
4.8 Overfitting, Underfitting & Bias-Variance Tradeoff
The most important concept in ML practice. Why models that memorise training data fail on new data. Regularisation techniques (L1/L2), cross-validation, and train/test/validation splits.
4.9 Model Evaluation Metrics
Accuracy, Precision, Recall, F1-Score, ROC-AUC for classification; MSE, RMSE, MAE, Rยฒ for regression. Choosing the right metric for your problem and business context.
4.10 Feature Engineering & Selection
Transforming raw data into meaningful inputs for models. One-hot encoding, normalisation, feature scaling, handling missing values, and dimensionality reduction with PCA.
4.11 Scikit-learn โ ML in Practice
Python’s premier ML library. Training, evaluating, and tuning models with consistent, beginner-friendly API. Pipelines, GridSearchCV, and cross-validation workflows.
05
Chapter Five ยท Core Concepts
Deep Learning & Neural Networks
Deep Learning is behind the AI revolution. Neural networks inspired by the human brain have achieved superhuman performance in vision, language, and strategy games โ and this chapter demystifies how they work.
How a Neural Network Learns
Forward Pass โ Loss Calculation โ Backpropagation โ Weight Update
Topics in this Chapter
5.1 The Perceptron โ The Simplest Neural Unit
How a single artificial neuron takes weighted inputs, applies a bias, and passes through an activation function. The building block of all neural networks, explained from first principles.
5.2 Multi-Layer Perceptrons (MLPs)
Stacking layers of neurons to model complex non-linear relationships. Understanding input, hidden, and output layers, and why depth (many layers) enables solving harder problems.
5.3 Activation Functions
ReLU, Sigmoid, Tanh, Softmax โ why non-linear activation functions are essential for neural networks to learn complex patterns. Comparing them and knowing when to use each.
5.4 Forward Propagation
Step-by-step walkthrough of how input data flows through a network to produce a prediction. Matrix multiplication at every layer, and how the final output is generated.
5.5 Backpropagation & The Chain Rule
How networks compute the gradient of the loss with respect to every weight using the chain rule of calculus. The algorithm that enables neural networks to learn โ explained intuitively.
5.6 Optimisers โ Adam, SGD, RMSProp
Algorithms that update network weights during training. Comparing Stochastic Gradient Descent, Momentum, RMSProp, and Adam โ the default choice for most modern networks.
5.7 Convolutional Neural Networks (CNNs)
The architecture behind image recognition. Convolution operations, filters, pooling layers, and how CNNs automatically detect edges, shapes, and objects in images.
5.8 Recurrent Neural Networks (RNNs) & LSTMs
Networks with memory โ designed for sequential data like text and time series. The vanishing gradient problem, and how LSTMs solve it with gating mechanisms.
5.9 The Transformer Architecture
The revolutionary architecture behind GPT, BERT, and all modern large language models. Self-attention, multi-head attention, positional encoding, and why Transformers replaced RNNs.
5.10 Regularisation โ Dropout, Batch Normalisation
Techniques to prevent overfitting in deep networks. Dropout randomly disables neurons during training; Batch Norm normalises layer inputs for stable, fast training.
5.11 Transfer Learning & Fine-Tuning
Using a pre-trained model as a starting point for a new task. Why training from scratch is rarely necessary, and how to fine-tune models with small datasets for domain-specific problems.
5.12 Training Deep Networks โ Practical Tips
Learning rate schedulers, weight initialisation, gradient clipping, early stopping, and monitoring training with TensorBoard. The difference between a model that trains and one that doesn’t.
06
Chapter Six ยท Core Concepts
Natural Language Processing
Teaching machines to understand human language is one of AI’s greatest challenges โ and greatest achievements. This chapter covers the concepts behind text AI, from classic methods to modern large language models.
Topics in this Chapter
6.1 What is NLP?
The field of AI focused on human language understanding. Key tasks: text classification, sentiment analysis, machine translation, summarisation, question answering, and text generation.
6.2 Text Preprocessing
Tokenisation, lowercasing, stopword removal, stemming, lemmatisation, and punctuation handling. Transforming raw text into clean inputs that models can process.
6.3 Text Representations โ Bag of Words & TF-IDF
Converting text to numbers using frequency-based methods. Bag of Words treats text as unordered word counts; TF-IDF weighs words by how unique they are across documents.
6.4 Word Embeddings โ Word2Vec, GloVe
Dense vector representations that capture semantic meaning. “king – man + woman โ queen” โ how word embeddings encode relationships and why they dramatically improved NLP models.
6.5 Attention Mechanism & Self-Attention
How models learn to focus on relevant parts of a sentence. The breakthrough behind Transformers โ understanding attention weights and query-key-value operations from first principles.
6.6 BERT โ Bidirectional Language Understanding
Google’s pre-trained model that reads text in both directions for context. Pre-training with masked language modelling, fine-tuning for downstream tasks, and its impact on NLP benchmarks.
6.7 GPT & Generative Language Models
Autoregressive language models that predict the next token. From GPT-1 to GPT-4 โ understanding how language models are trained, how prompting works, and the concept of emergent abilities.
6.8 Sentiment Analysis โ A Practical NLP Task
Building a classifier that determines if text is positive, negative, or neutral. End-to-end walkthrough using both traditional ML and transformer-based models.
6.9 Named Entity Recognition (NER)
Identifying and classifying entities in text โ people, organisations, locations, dates. A foundational NLP task used in information extraction and document understanding pipelines.
6.10 Using Hugging Face Transformers
The most important NLP library in practice. Loading pre-trained models, tokenising text, running inference, and fine-tuning BERT or GPT-2 on your own dataset in under 50 lines of code.
07
Chapter Seven ยท Specialisation
Computer Vision
Computer Vision enables machines to see and understand the visual world. From medical imaging to autonomous vehicles, it is one of AI’s most impactful โ and visually intuitive โ fields.
Topics in this Chapter
7.1 Introduction to Computer Vision
How images are represented as pixel grids and numerical arrays. Colour channels (RGB), image resolution, and the challenge of teaching machines to interpret visual data.
7.2 Image Classification
Assigning a label to an entire image โ “cat” or “dog.” The classic entry point to computer vision. Using CNNs and pre-trained models like ResNet and VGG on standard datasets.
7.3 Object Detection
Not just what is in an image, but where. Bounding boxes, anchor boxes, and landmark architectures: YOLO (You Only Look Once), Faster R-CNN, and SSD for real-time detection.
7.4 Image Segmentation
Pixel-level understanding of images โ semantic segmentation assigns every pixel a class, while instance segmentation distinguishes individual objects of the same class.
7.5 Convolutional Filters & Feature Maps
Deep dive into how CNNs extract features. Visualising what different layers learn: early layers detect edges, middle layers find shapes, deep layers recognise complex objects.
7.6 Data Augmentation for Images
Artificially expanding training datasets through flipping, rotation, cropping, colour jitter, and Cutout. Essential for training robust models when labelled data is scarce.
7.7 Transfer Learning with Pre-Trained Vision Models
Fine-tuning ImageNet-trained models (ResNet, EfficientNet, ViT) for custom tasks. How to freeze layers, replace the classification head, and achieve high accuracy with minimal data.
7.8 Generative Models โ GANs & Diffusion
Creating new images with AI. Generative Adversarial Networks (GANs) pit a generator against a discriminator. Diffusion models (Stable Diffusion, DALLยทE) reverse a noise process to generate photorealistic images.
7.9 OpenCV โ Computer Vision in Practice
The most widely used computer vision library. Image loading, resizing, filtering, edge detection with Canny, contour detection, and real-time video processing with your webcam.
08
Chapter Eight ยท Specialisation
AI Tools, Frameworks & Ecosystem
The right tools can cut weeks of work to hours. This chapter surveys the essential AI software ecosystem โ the frameworks, platforms, and services every AI practitioner should know.
Framework Comparison
| Framework |
Best For |
Difficulty |
Key Feature |
| TensorFlow / Keras |
Production & deployment |
Intermediate |
Keras high-level API, TFLite for mobile |
| PyTorch |
Research & experimentation |
Intermediate |
Dynamic graphs, Pythonic feel |
| Scikit-learn |
Classical ML |
Beginner |
Consistent API across all algorithms |
| Hugging Face |
NLP & Transformers |
BeginnerโIntermediate |
10,000+ pre-trained models |
| JAX |
Research, TPUs |
Advanced |
Functional transformations, GPU/TPU |
Topics in this Chapter
8.1 TensorFlow & Keras
Google’s flagship AI framework. Building and training neural networks with Keras’s beginner-friendly API. The Sequential and Functional model APIs, saving/loading models, and TensorBoard for visualisation.
8.2 PyTorch
Facebook’s research-oriented framework and the dominant choice in AI research papers. Dynamic computation graphs, custom training loops, autograd, and DataLoaders for efficient batching.
8.3 Hugging Face Ecosystem
Transformers, Datasets, Accelerate, and the Model Hub. The de facto standard for working with pre-trained language models. Running inference, fine-tuning, and sharing models with the community.
8.4 LangChain & LLM Application Frameworks
Building applications on top of large language models. Chains, agents, memory, and retrieval-augmented generation (RAG). Creating chatbots, document Q&A systems, and autonomous agents.
8.5 Cloud AI Platforms
Google Vertex AI, AWS SageMaker, Azure ML โ managed platforms for training, deploying, and monitoring models at scale without managing infrastructure. Understanding when cloud vs local training makes sense.
8.6 Experiment Tracking โ MLflow & Weights & Biases
Logging hyperparameters, metrics, and model artefacts across experiments. Comparing runs, visualising learning curves, and reproducing the best experiments โ professional ML workflow practices.
8.7 Model Deployment โ From Notebook to Production
Serving AI models as APIs using FastAPI or Flask. Docker containerisation, REST endpoint design, input/output schemas, and monitoring deployed models for data drift and performance degradation.
8.8 Version Control for AI โ Git & DVC
Managing code with Git and tracking datasets and model versions with DVC (Data Version Control). Reproducible AI pipelines and collaborative workflows for AI teams.
10
Chapter Ten ยท Real World
AI Ethics, Responsibility & The Future
As AI becomes more powerful, the responsibility of those who build it grows. Every AI practitioner must understand bias, fairness, safety, and the societal impact of the systems they create.
Topics in this Chapter
10.1 Bias in AI Systems
How historical biases in training data get encoded into models โ and amplified. Facial recognition bias, hiring algorithm discrimination, and loan approval disparities. Understanding sources of bias and how to measure it.
10.2 Fairness, Accountability & Transparency
The FAT framework for responsible AI. Multiple definitions of algorithmic fairness (demographic parity, equal opportunity) and why they can conflict. Auditability and the right to explanation.
10.3 Explainable AI (XAI)
Making AI decisions interpretable. LIME and SHAP for explaining model predictions, attention visualisation, and why “black box” models are problematic in high-stakes domains like medicine and law.
10.4 AI Safety & Alignment
Ensuring AI systems do what we intend. Specification gaming, reward hacking, and the challenge of aligning powerful models with human values. Introduction to RLHF (Reinforcement Learning from Human Feedback).
10.5 Privacy & Data Ethics
Federated learning, differential privacy, and GDPR compliance. How AI systems can be trained without exposing sensitive personal data. Data collection ethics and informed consent.
10.6 Environmental Impact of AI
The carbon cost of training large models โ GPT-3 training emitted as much COโ as five cars over their lifetimes. Efficient model design, model distillation, and the responsibility to measure and reduce AI’s energy footprint.
10.7 Deepfakes & Generative AI Risks
How generative models can be misused for misinformation, fraud, and manipulation. Detection methods, watermarking, and the policy landscape around synthetic media.
10.8 AI Regulation & Global Policy
The EU AI Act, US AI Executive Orders, and China’s AI governance frameworks. How regulations classify AI risk, and what compliance means for developers and companies building AI products.
10.9 The Future of AI โ Trends & Opportunities
Multimodal models, autonomous AI agents, AI in science (AlphaFold, climate modelling), brain-computer interfaces, and the potential path toward Artificial General Intelligence. Emerging roles for AI practitioners.
10.10 Your Next Steps as an AI Learner
Building your portfolio, contributing to open-source AI projects, participating in Kaggle competitions, reading research papers, and navigating AI career paths โ data scientist, ML engineer, AI researcher, and beyond.
“The question is not whether AI will change the world, but whether the people building it will do so responsibly.”
โ AI Ethics Principle, Anthropic 2024
๐ Congratulations โ You’ve Covered It All
Completing this course material gives you a comprehensive foundation in AI. You now understand how machines learn, the mathematics behind neural networks, the tools practitioners use, real-world applications, and the ethical responsibilities that come with building AI systems. The next step is practice โ start a project, enter a competition, and keep building.
Recommended Learning Resources
04
Kaggle LearnFree micro-courses on Python, ML, deep learning, NLP, and data visualisation with hands-on notebooks.
06
Papers with CodeLatest AI research papers paired with implementation code. The best way to stay current with the field.
