AI & Machine Learning
A comprehensive, 3-pillar curriculum that builds durable understanding of artificial intelligence and machine learning — from the mathematical foundations and classical algorithms, through modern deep learning architectures, to the societal implications of deploying AI in the real world. Emphasizes timeless principles over transient tools so knowledge remains valuable as the field evolves.
670 lessons · about 1000 hours
What you'll learn
Foundations of Intelligent Systems
The mathematical and algorithmic bedrock of AI and machine learning. Covers the math you need (linear algebra, calculus, probability, statistics), classical ML algorithms, optimization, and probabilistic reasoning. These principles are timeless — they haven't changed in decades and won't change in the next decade.
Unit 1
Mathematical Foundations for Machine Learning
Build the essential mathematical toolkit required for understanding machine learning algorithms. This course covers linear algebra, calculus, and core mathematical concepts that form the backbone of every ML technique, from simple linear regression to deep neural networks.
22 lessons
- Vectors and Scalars
- Matrix Operations
- Eigenvalues and Eigenvectors
- Derivatives and Rates of Change
- Inner Products and Their Properties
- Dot Product and Projections
- Linear Transformations
- Eigendecomposition
- Partial Derivatives and Gradients
- Orthogonality and Orthogonal Projections
- Norms and Distance Measures
- Determinants and Invertibility
- Singular Value Decomposition (SVD)
- The Chain Rule
- The Gram-Schmidt Process
- Linear Independence and Basis
- Systems of Linear Equations
- Positive Definite and Semi-Definite Matrices
- Taylor Series Approximation
- From Inner Products to the Kernel Trick
- Vector Spaces and Subspaces
- Jacobian and Hessian Matrices
Unit 2
What Is Artificial Intelligence
A comprehensive introduction to the field of artificial intelligence, tracing its origins from the philosophical questions about machine thought through the landmark achievements that have shaped the modern AI landscape. This course covers the key definitions, paradigms, and debates that every AI practitioner should understand.
22 lessons
- Early Automata and Mechanical Thought
- Defining Intelligence
- Narrow AI in Practice
- Deep Blue and Chess
- The Turing Test
- Turing's Foundational Contributions
- Symbolic AI (GOFAI)
- Artificial General Intelligence (AGI)
- Watson and Jeopardy!
- The Chinese Room Argument
- The Dartmouth Conference and the Birth of AI
- Statistical and Connectionist AI
- Superintelligence and Existential Risk
- AlphaGo and the Game of Go
- Consciousness and Machines
- AI Winters and Hype Cycles
- Hybrid and Neurosymbolic Approaches
- Measuring Progress Toward AGI
- GPT and the Large Language Model Revolution
- Ethics and Responsibility in AI
- The Modern AI Resurgence
- AlphaFold and Scientific Discovery
Unit 3
Model Evaluation and Selection
Master the tools and techniques for rigorously evaluating machine learning models and selecting the best model for a given problem. This course covers evaluation metrics, cross-validation strategies, hyperparameter tuning methods, and the statistical considerations needed to draw reliable conclusions about model performance.
22 lessons
- The Confusion Matrix
- MSE, RMSE, and MAE
- K-Fold Cross-Validation
- Parameters vs. Hyperparameters
- Statistical Significance of Model Differences
- Accuracy and Its Limitations
- R-Squared and Adjusted R-Squared
- Stratified and Grouped Cross-Validation
- Grid Search
- Probability Calibration
- Precision, Recall, and F1 Score
- MAPE and Relative Error Metrics
- Time Series Cross-Validation
- Random Search
- Learning Curves as Diagnostic Tools
- ROC Curve and AUC
- Residual Analysis
- Nested Cross-Validation
- Bayesian Optimization
- Systematic Model Selection Workflow
- Precision-Recall Curve and Average Precision
- Early Stopping and Successive Halving
Unit 4
How Machines Learn
Understand the fundamental principles behind machine learning at a conceptual level, including the major learning paradigms, the role of loss functions and optimization, and the critical tradeoffs that govern every ML system. This course bridges the gap between knowing what AI is and understanding how it actually works.
21 lessons
- Supervised Learning
- Hypothesis Space
- Loss Functions
- Bias-Variance Decomposition
- Regularization Intuition
- Unsupervised Learning
- Inductive Bias
- Empirical Risk Minimization
- Overfitting
- L1 and L2 Regularization
- Reinforcement Learning
- Model Complexity
- Structural Risk Minimization
- Underfitting
- Early Stopping and Dropout
- Self-Supervised Learning
- Representation and Feature Spaces
- Surrogate Loss Functions
- Learning Curves
- The No Free Lunch Theorem
- Double Descent and Modern Interpolation
Unit 5
Supervised Learning
A comprehensive tour of the most important supervised learning algorithms, from classical linear models to powerful ensemble methods. This course covers the mathematical intuition, practical strengths and weaknesses, and appropriate use cases for each algorithm, preparing you to select and apply the right model for any supervised learning problem.
22 lessons
- Linear Regression
- Decision Trees
- K-Nearest Neighbors (k-NN)
- Maximum Margin Classifier
- Why Ensembles Work
- Polynomial and Basis Function Expansion
- Tree Pruning and Complexity Control
- The Curse of Dimensionality
- Support Vectors and Soft Margin
- Bagging and Pasting
- Logistic Regression
- Random Forests
- Naive Bayes Classifier
- Kernel SVM
- Boosting Principles: AdaBoost and Beyond
- Regularized Linear Models: Ridge, Lasso, and Elastic Net
- Gradient Boosting
- Choosing k and Distance Metrics
- SVM for Regression (SVR)
- Stacking (Stacked Generalization)
- Multiclass Extensions
- Feature Importance in Tree-Based Models
Unit 6
Probabilistic Graphical Models
Explore the powerful framework of probabilistic graphical models that combine probability theory with graph theory to represent and reason about complex systems with many interacting variables. This advanced course covers Bayesian networks, Markov random fields, inference algorithms, and their applications to sequence modeling and causal reasoning.
20 lessons
- Directed Graphical Models
- Undirected Graphical Models (MRFs)
- Variable Elimination
- Monte Carlo Sampling Methods
- Hidden Markov Models (HMMs)
- Conditional Independence and D-Separation
- Directed vs. Undirected Models
- Belief Propagation (Sum-Product Algorithm)
- Markov Chain Monte Carlo (MCMC)
- Forward-Backward and Viterbi Algorithms
- Parameter Learning in Bayesian Networks
- Factor Graphs
- The Junction Tree Algorithm
- Variational Inference
- Conditional Random Fields (CRFs)
- Structure Learning
- Energy-Based Models
- Computational Complexity of Inference
- Loopy Belief Propagation
- Causal Inference with Graphical Models
Unit 7
Unsupervised Learning
Explore algorithms that discover hidden structure in unlabeled data, from clustering methods that group similar observations to dimensionality reduction techniques that reveal the essential structure of high-dimensional datasets. This course covers both classical techniques and modern visualization methods.
20 lessons
- The K-Means Algorithm
- Agglomerative Hierarchical Clustering
- Gaussian Mixture Models (GMMs)
- Principal Component Analysis (PCA)
- Anomaly Detection Approaches
- Initialization Strategies: K-Means++
- Linkage Criteria and Their Effects
- The Expectation-Maximization (EM) Algorithm
- Explained Variance and Choosing Components
- Isolation Forest
- Choosing the Number of Clusters
- DBSCAN: Density-Based Clustering
- GMM Model Selection: BIC and AIC
- t-SNE: t-Distributed Stochastic Neighbor Embedding
- Association Rule Mining
- K-Medoids and K-Means Variants
- HDBSCAN: Hierarchical Density-Based Clustering
- Covariance Types and Their Impact
- UMAP: Uniform Manifold Approximation and Projection
- Evaluating Unsupervised Learning
Unit 8
Probability and Statistics for Machine Learning
Develop a rigorous understanding of probability theory and statistical methods that underpin machine learning algorithms. From Bayes' theorem to information theory, this course equips you with the probabilistic reasoning skills essential for understanding generative models, Bayesian inference, and model evaluation.
23 lessons
- Probability Axioms and Sample Spaces
- Random Variables and Their Types
- Maximum Likelihood Estimation (MLE)
- Null and Alternative Hypotheses
- Entropy
- Conditional Probability
- Expectation & Variance
- Maximum A Posteriori Estimation (MAP)
- P-Values and Statistical Significance
- Cross-Entropy
- Law of Total Probability
- Bernoulli & Binomial Distributions
- Bias and Variance of Estimators
- Confidence Intervals
- KL Divergence
- Independence and Conditional Independence
- The Gaussian (Normal) Distribution
- Consistency and Efficiency
- Type I and Type II Errors
- Mutual Information
- Bayes' Theorem
- Other Key Distributions: Poisson, Exponential, and Categorical
- Multiple Testing and Correction Methods
Unit 9
Data Thinking and Preparation
Learn the foundational skills of working with data that determine the success or failure of any machine learning project. This course covers data types, collection strategies, cleaning techniques, feature engineering, and the critical practices that prevent data leakage and ensure reliable model evaluation.
20 lessons
- Structured vs. Unstructured Data
- Data Collection Methods
- Missing Data Mechanisms
- Exploratory Data Analysis (EDA)
- Train, Validation, and Test Splits
- Numerical and Categorical Features
- Sampling Bias
- Imputation Strategies
- Normalization and Standardization
- Stratified and Temporal Splitting
- Ordinal Data
- Label Quality and Annotation
- Outlier Detection and Treatment
- Feature Engineering Basics
- Data Leakage
- Time Series Data
- Ethical Data Practices
- Deduplication and Data Consistency
- Encoding Categorical Features
- Leakage-Prevention Pipelines
Unit 10
Optimization and Search
Explore the optimization algorithms and search strategies that enable machines to learn from data and solve complex problems. From gradient descent, the workhorse of modern ML, to classical search algorithms and metaheuristic approaches, this course covers the computational methods that find good solutions in vast search spaces.
21 lessons
- Gradient Descent Fundamentals
- Convex Functions
- Constrained Optimization Formulations
- Breadth-First and Depth-First Search
- Simulated Annealing
- Batch, Mini-Batch, and Stochastic Gradient Descent
- Non-Convex Optimization in Deep Learning
- Lagrange Multipliers
- A* Search
- Genetic Algorithms
- The Learning Rate
- Saddle Points and Plateaus
- KKT Conditions
- Heuristic Design and Admissibility
- Particle Swarm Optimization
- Momentum and Nesterov Acceleration
- Convergence Conditions and Guarantees
- Primal-Dual Formulations
- Iterative Deepening and Bidirectional Search
- When to Use Metaheuristics
- Adaptive Learning Rate Methods: Adam, RMSProp, AdaGrad
Deep Learning & Modern Architectures
Neural networks from first principles through modern architectures — CNNs, transformers, generative models, reinforcement learning, and graph neural networks. Taught as mathematics and architecture rather than framework tutorials, so the knowledge survives tool churn.
Unit 11
Neural Networks from Scratch
Build a deep understanding of neural networks by constructing them from first principles. Starting with the biological inspiration behind artificial neurons, you will derive the mathematics of the perceptron, compose multi-layer networks, implement forward propagation and backpropagation, and understand why neural networks are universal function approximators. This course bridges the gap between mathematical foundations and modern deep learning frameworks.
25 lessons
- Biological Neurons and Artificial Inspiration
- The Role of Nonlinearity
- Forward Propagation Mechanics
- Chain Rule for Composite Functions
- Universal Approximation Theorem
- The Perceptron Model
- Sigmoid Activation Function
- Mean Squared Error Loss
- Backpropagation Derivation
- Depth vs Width Tradeoffs
- Perceptron Learning Rule
- Tanh Activation Function
- Cross-Entropy Loss
- Computational Graphs
- Why Initialization Matters
- Limitations and the XOR Problem
- ReLU and Its Variants
- Loss Surfaces and Optimization Landscape
- Gradient Flow and Debugging
- Xavier (Glorot) Initialization
- From Single Neurons to Multi-Layer Networks
- Softmax Activation Function
- Choosing and Combining Loss Functions
- Automatic Differentiation in Practice
- He (Kaiming) Initialization
Unit 12
Reinforcement Learning
Learn how agents learn optimal behavior through interaction with environments. This course covers the mathematical framework of Markov decision processes, progresses through dynamic programming, Monte Carlo, and temporal difference methods, and culminates in modern deep reinforcement learning algorithms. You will understand the exploration-exploitation tradeoff, policy gradient methods, and the actor-critic architectures that have achieved superhuman performance in games and robotics.
25 lessons
- Agent-Environment Interface
- Dynamic Programming for MDPs
- TD Prediction (TD(0))
- Policy Gradient Theorem
- Deep Q-Networks (DQN)
- Markov Decision Processes
- Value Iteration
- SARSA: On-Policy TD Control
- REINFORCE Algorithm
- Double DQN and Dueling DQN
- Value Functions
- Monte Carlo Prediction
- Q-Learning: Off-Policy TD Control
- Variance Reduction with Baselines
- Proximal Policy Optimization (PPO)
- Bellman Equations
- Exploration vs Exploitation
- TD(lambda) and Eligibility Traces
- Actor-Critic Architecture
- RL in Continuous Action Spaces
- Policies and Optimality
- Monte Carlo Control
- Reward Shaping
- Generalized Advantage Estimation (GAE)
- Challenges and Frontiers of Deep RL
Unit 13
Graph and Geometric Machine Learning
Extend deep learning beyond grids and sequences to the rich world of graphs and geometric structures. This course covers how to apply neural networks to irregularly structured data like social networks, molecular graphs, and point clouds. Starting from graph theory fundamentals for ML, you will progress through message passing, graph neural network architectures, and the geometric deep learning framework that unifies CNNs, transformers, and GNNs under the principles of symmetry and equivariance.
23 lessons
- Graph Fundamentals for Machine Learning
- Message Passing Framework
- Graph Convolutional Networks (GCN)
- Molecular Graphs and Drug Discovery
- Equivariance and Invariance
- Graph Representations for Neural Networks
- Aggregation Functions
- Graph Attention Networks (GAT)
- Social Network Analysis with GNNs
- Symmetry Groups in Machine Learning
- Graph Laplacian and Spectral Properties
- Multi-Layer Information Propagation
- Node, Edge, and Graph-Level Tasks
- Graph-Based Recommendation Systems
- Unifying CNNs, GNNs, and Transformers
- Challenges of Learning on Graphs
- Over-Smoothing Problem
- Graph Pooling and Readout Functions
- Point Clouds and 3D Data
- Equivariant Neural Networks
- Expressiveness and the WL Test
- Knowledge Graph Reasoning
- Frontiers of Geometric Machine Learning
Unit 14
Sequence Models and Attention
Chart the evolution from recurrent neural networks to the transformer architecture that powers modern AI. Beginning with the fundamental challenge of modeling sequential data, this course progresses through RNNs, LSTMs, and GRUs, then pivots to the attention mechanism and the transformer architecture that replaced recurrence with parallelizable self-attention. You will understand the design decisions behind the most influential neural architecture of the current era.
24 lessons
- RNN Fundamentals
- LSTM Architecture
- Attention Intuition and Motivation
- Positional Encoding
- Scaling Laws for Neural Language Models
- Backpropagation Through Time
- LSTM Gate Mechanics
- Query-Key-Value Framework
- Encoder-Decoder Transformer
- Encoder-Only Models (BERT Paradigm)
- Sequence-to-Sequence Framework
- GRU Architecture
- Scaled Dot-Product Attention
- Causal Masking
- Decoder-Only Models (GPT Paradigm)
- Teacher Forcing
- LSTM vs GRU: When to Use Which
- Self-Attention
- Feed-Forward Networks in Transformers
- Encoder-Decoder Models (T5 Paradigm)
- Bidirectional RNNs
- Multi-Head Attention
- Residual Connections and Layer Normalization in Transformers
- Efficient Attention Variants
Unit 15
Training Deep Networks
Master the techniques that make training deep neural networks practical and reliable. From diagnosing and fixing gradient pathologies to selecting optimizers, normalization strategies, and regularization methods, this course covers the engineering craft behind successful deep learning. You will learn to tune learning rate schedules, apply mixed precision training for efficiency, and leverage transfer learning to bootstrap new models from pretrained ones.
24 lessons
- Vanishing Gradients
- Internal Covariate Shift
- Overfitting in Deep Networks
- SGD with Momentum
- Mixed Precision Training Fundamentals
- Exploding Gradients
- Batch Normalization
- Dropout Regularization
- AdaGrad and RMSProp
- Loss Scaling for Mixed Precision
- Gradient Clipping
- Layer Normalization
- Early Stopping
- Adam Optimizer
- Transfer Learning Fundamentals
- Diagnosing Gradient and Activation Issues
- Group and Instance Normalization
- Weight Decay and L2 Regularization
- Learning Rate Warmup
- Fine-Tuning Strategies
- Normalization Placement Strategies
- Data Augmentation as Regularization
- Cosine Annealing and Other Schedules
- Feature Extraction vs Full Fine-Tuning
Unit 16
ML Engineering and Operations
Bridge the gap between training models and running them reliably in production. This course covers the full machine learning lifecycle from experiment tracking and reproducibility through deployment, monitoring, and maintenance. You will learn the engineering practices, tools, and system design patterns that enable teams to develop, deploy, and iterate on ML systems with confidence and at speed.
23 lessons
- The ML Lifecycle
- Data Versioning
- ML Testing Pyramid
- Model Monitoring Fundamentals
- ML System Design Patterns
- Experiment Tracking
- Model Registry and Versioning
- Continuous Training Pipelines
- Data Drift Detection
- Technical Debt in ML Systems
- Reproducibility in ML
- Feature Stores
- Model Validation Gates
- Concept Drift
- Data Management Best Practices
- Hyperparameter Management
- Data and Model Lineage
- ML Deployment Strategies
- A/B Testing for ML Models
- ML Governance and Documentation
- Rollback and Recovery
- Alerting and Incident Response
- ML Team Practices and Collaboration
Unit 17
Scalable ML Systems
Learn how to train and deploy machine learning models at scale. This course covers the hardware, parallelism strategies, and optimization techniques that make it possible to train models with billions of parameters and serve them efficiently in production. From data parallelism and model sharding to quantization and knowledge distillation, you will understand the systems engineering that turns research models into practical, deployable AI systems.
23 lessons
- GPU Architecture for Deep Learning
- Data Parallelism
- Anatomy of GPU Memory Usage
- Model Pruning
- Batch Serving vs Real-Time Serving
- TPU Architecture and Comparison
- Model Parallelism
- Activation Checkpointing (Gradient Checkpointing)
- Model Quantization
- Inference Optimization Techniques
- Memory and Bandwidth Bottlenecks
- Pipeline Parallelism
- ZeRO: Zero Redundancy Optimizer
- Knowledge Distillation
- Speculative Decoding
- Hardware Selection for ML Workloads
- Gradient Accumulation
- CPU and NVMe Offloading
- Compression Tradeoffs and Combining Techniques
- ML Serving Infrastructure
- 3D Parallelism and Hybrid Strategies
- Flash Attention and Memory-Efficient Attention
- Cost Optimization for ML Inference
Unit 18
Convolutional Neural Networks
Explore the architecture that revolutionized computer vision. Starting from the convolution operation and its properties, this course builds up through pooling, feature maps, and the landmark architectures that pushed the boundaries of image recognition. You will understand how CNNs exploit spatial structure to learn hierarchical visual features, and master the design principles behind modern convolutional architectures.
20 lessons
- Convolution Fundamentals
- Max Pooling
- LeNet: The Pioneer
- 1x1 Convolutions
- Filters and Kernels
- Average and Global Pooling
- AlexNet: The Deep Learning Revolution
- Depthwise Separable Convolutions
- Stride and Padding
- Feature Map Hierarchy
- VGGNet: Depth with Simplicity
- Dilated (Atrous) Convolutions
- Output Dimension Calculation
- Receptive Field
- ResNet: Skip Connections and Depth
- Transposed Convolutions
- Translation Equivariance and Weight Sharing
- Channels and Depth Dimension
- Architecture Evolution Principles
- CNN Architectural Design Principles
Unit 19
Generative Models
Explore the architectures that learn to generate new data — images, text, audio, and beyond. This course covers the fundamental distinction between generative and discriminative modeling, then dives into autoencoders, variational autoencoders, generative adversarial networks, diffusion models, and autoregressive generation. You will understand the mathematical foundations, training dynamics, and evaluation metrics that define each approach.
24 lessons
- Generative vs Discriminative Models
- Basic Autoencoders
- GAN Framework and Minimax Game
- Diffusion Model Intuition
- Frechet Inception Distance (FID)
- Modeling High-Dimensional Distributions
- Variational Autoencoders (VAE)
- Mode Collapse and Training Instability
- Denoising Score Matching
- Inception Score (IS)
- Latent Variable Models
- Evidence Lower Bound (ELBO)
- Wasserstein GAN (WGAN)
- Normalizing Flows
- Quality vs Diversity Tradeoff
- Maximum Likelihood for Generative Models
- The Reparameterization Trick
- Conditional GANs
- Classifier-Free Guidance
- Comparing Generative Model Families
- Latent Space Properties and Interpolation
- Progressive Training and StyleGAN
- Autoregressive Generative Models
- Applications of Generative Models
Unit 20
Representation Learning
Understand how neural networks learn useful representations of data, from static word embeddings to the self-supervised paradigms that power modern foundation models. This course covers the progression from hand-crafted features to learned representations, the techniques for embedding discrete and continuous data into meaningful vector spaces, and the contrastive and predictive objectives that enable learning from unlabeled data at scale.
23 lessons
- From Hand-Crafted Features to Learned Representations
- Contextual Embeddings
- Self-Supervised Learning Paradigm
- What is Disentanglement?
- Transfer Learning Theory and Practice
- Embedding Spaces
- Masked Language Modeling
- Contrastive Learning Framework
- Beta-VAE and Regularized Disentanglement
- Domain Adaptation
- Word2Vec
- Next-Token Prediction as Representation Learning
- SimCLR: Simple Contrastive Learning
- Measuring Disentanglement
- Foundation Models
- GloVe Embeddings
- Probing and Understanding Representations
- CLIP: Contrastive Language-Image Pretraining
- Theoretical Limits of Unsupervised Disentanglement
- Parameter-Efficient Fine-Tuning
- Limitations of Static Embeddings
- Non-Contrastive Self-Supervised Methods
- Risks and Limitations of Foundation Models
AI in the Real World
Applying AI to real domains (NLP, vision, recommender systems) and reasoning about its societal impact — safety, alignment, fairness, governance, human-AI interaction, and economics. These concerns only grow more important as AI capabilities increase.
Unit 21
AI Governance & Regulation
Navigate the rapidly evolving landscape of AI governance, from landmark regulations like the EU AI Act to corporate responsible AI frameworks. This course examines how governments, international bodies, and industry are establishing rules for AI development and deployment, covering risk-based regulatory approaches, technical standards, intellectual property challenges, data privacy intersections, and the open source debate. You will develop the analytical skills to evaluate governance proposals and understand their implications for AI practitioners.
21 lessons
- The EU AI Act
- Risk-Based AI Governance Frameworks
- Responsible AI Frameworks
- Copyright & AI Training Data
- Open Source vs Closed Models Debate
- US AI Policy & Executive Orders
- ISO & IEEE AI Standards
- AI Governance Organizational Structure
- Ownership of AI-Generated Content
- Foundation Model Governance
- China's AI Regulations
- AI Risk Assessment in Practice
- AI Lifecycle Governance
- GDPR & Machine Learning
- International AI Governance Cooperation
- Regulatory Approaches Compared
- Conformity Assessment & Certification
- AI Incident Response
- Privacy-Preserving ML Techniques
- Compute Governance
- AI & Patent Law
Unit 22
AI Safety & Alignment
Confront the fundamental challenge of building AI systems that reliably do what we intend. This course covers the alignment problem from multiple angles: specification challenges like reward hacking and specification gaming, training-time alignment techniques including RLHF and constitutional AI, post-hoc interpretability methods, adversarial robustness, and the broader debate around existential risk. You will develop the conceptual toolkit to reason about AI safety as both a current engineering challenge and a long-term research frontier.
23 lessons
- The Alignment Problem Defined
- Reinforcement Learning from Human Feedback (RLHF)
- LIME (Local Interpretable Model-Agnostic Explanations)
- Adversarial Examples
- Existential Risk Arguments
- Reward Hacking
- Constitutional AI
- SHAP (SHapley Additive exPlanations)
- Adversarial Attack Methods
- AI Containment & Boxing
- Specification Gaming
- AI Debate & Amplification
- Attention Visualization
- Adversarial Defense Strategies
- Corrigibility & Shutdown Problem
- Outer vs Inner Alignment
- Limitations of Reward Modeling
- Mechanistic Interpretability
- Distribution Shift & Out-of-Distribution Detection
- The Bridge from Current to Future Safety
- Mesa-Optimization
- Faithfulness vs Usefulness of Explanations
- Responsible AI Development Practices
Unit 23
AI Economics & Labor
Analyze the profound economic transformations driven by artificial intelligence, from labor market disruption and productivity gains to the concentration of economic power and the global distribution of AI's benefits and costs. This course examines how AI reshapes work, creates and destroys jobs, alters competitive dynamics, and challenges existing economic institutions. You will develop the economic literacy to reason about AI's impact on employment, inequality, and prosperity.
20 lessons
- AI in Developing Economies
- Historical Patterns of Technological Unemployment
- AI Productivity Gains
- Economics of Training Large Models
- AI & Economic Inequality
- AI Geopolitics & Competition
- Task-Based Framework for Automation
- Comparative Advantage with AI
- Data as an Economic Asset
- Universal Basic Income Debate
- The AI Digital Divide
- Evidence on AI Job Displacement
- The AI Productivity Paradox
- Winner-Take-All Dynamics
- Retraining & Education Policy
- AI Taxation Strategies
- The Future of Work with AI
- AI-Driven Job Creation
- Skill Premiums & Job Polarization
- AI Business Models
Unit 24
Computer Vision Applications
Discover how convolutional and transformer-based architectures are applied to real-world visual understanding tasks. This course covers the major application areas of computer vision including image classification at scale, object detection, semantic and instance segmentation, generative image models, video understanding, 3D vision, and domain-specific applications in medical imaging and autonomous driving. You will learn the architectures, evaluation protocols, and deployment considerations that define production computer vision systems.
23 lessons
- Modern Classification Architectures
- Two-Stage Detectors (R-CNN Family)
- Semantic Segmentation
- Image Generation with GANs
- Medical Imaging Applications
- Transfer Learning for Vision
- Single-Stage Detectors (YOLO & SSD)
- Instance Segmentation
- Diffusion Models for Image Synthesis
- Autonomous Driving Perception
- Data Augmentation Strategies
- Anchor-Free Detection
- Panoptic Segmentation
- Neural Style Transfer
- Depth Estimation
- Classification Evaluation & Error Analysis
- Detection Evaluation Metrics
- Segmentation Loss Functions
- Video Understanding & Temporal Modeling
- Point Cloud Processing
- Vision Transformers
- Feature Pyramid Networks
- Object Tracking
Unit 25
Fairness, Accountability & Transparency
Develop a rigorous understanding of how AI systems can perpetuate or amplify societal biases and what technical and institutional mechanisms exist to address these harms. This course covers the mathematical definitions of fairness and their inherent tensions, methods for auditing algorithms, documentation practices like model cards and datasheets, and landmark case studies that illustrate how bias manifests in real-world AI deployments. You will learn to analyze, measure, and mitigate unfairness in machine learning systems.
23 lessons
- Sources of Bias in ML Pipelines
- Demographic Parity
- Audit Methodology & Frameworks
- Model Cards
- COMPAS & Criminal Justice Bias
- Allocative vs Representational Harms
- Equalized Odds & Equal Opportunity
- Disparate Impact Analysis
- Datasheets for Datasets
- Hiring Algorithm Bias
- Bias Amplification
- Individual vs Group Fairness
- Subgroup Performance Evaluation
- Algorithmic Impact Assessments
- Pre-Processing Bias Mitigation
- Dataset Bias Case Studies
- Fairness Impossibility Theorems
- External & Independent Auditing
- Transparency & Explainability Requirements
- In-Processing & Post-Processing Mitigation
- Choosing Fairness Metrics in Practice
- Open Source & Reproducibility
- Organizational Fairness Strategy
Unit 26
Recommender Systems
Learn how to build systems that predict what users want before they ask. This course covers the full landscape of recommendation algorithms from classical collaborative and content-based filtering through modern deep learning approaches, examining how platforms like Netflix, Amazon, and Spotify personalize content for billions of users. You will study matrix factorization, hybrid methods, evaluation metrics specific to ranking, exploration-exploitation trade-offs, and the ethical dimensions of algorithmic recommendation.
22 lessons
- User-Based Collaborative Filtering
- Content-Based Filtering
- Ranking Metrics (NDCG & MAP)
- Neural Collaborative Filtering
- Filter Bubbles & Echo Chambers
- Item-Based Collaborative Filtering
- Matrix Factorization
- Precision@K and Recall@K
- Sequential & Session-Based Recommendations
- Popularity Bias
- Similarity Measures for Recommendations
- Hybrid Recommendation Approaches
- Offline Evaluation Protocols
- Graph Neural Networks for Recommendations
- Multi-Stakeholder Fairness
- Implicit vs Explicit Feedback
- Factorization Machines
- Beyond-Accuracy Metrics
- Multi-Armed Bandits for Exploration
- Responsible Recommendation Design
- The Cold Start Problem
- A/B Testing Recommender Systems
Unit 27
Human-AI Interaction
Explore how humans and AI systems work together, focusing on the cognitive, social, and design factors that determine whether AI augments or undermines human capability. This course covers trust calibration, automation bias, human-AI teaming models, conversational AI design, accessibility considerations, and UX principles specific to AI-powered products. You will learn to design AI systems that are not just technically capable but genuinely usable and beneficial for the people who interact with them.
21 lessons
- Cognitive Biases in AI Interaction
- Augmentation vs Automation
- Types of AI Explanations
- Conversational Design Principles
- AI-Created Accessibility Barriers
- Trust Calibration
- Task Allocation in Human-AI Teams
- Communicating Uncertainty
- AI Persona & Tone Design
- AI as Assistive Technology
- Mental Models of AI
- Collaborative Intelligence
- When & How to Explain
- Error Recovery & Graceful Degradation
- Inclusive AI Design Principles
- Automation Complacency & Skill Degradation
- Adaptive Autonomy
- Explanation Pitfalls & Dark Patterns
- Cognitive Load in AI-Assisted Work
- AI UX Design Principles
- Multimodal AI Interaction
Unit 28
Frontiers & Open Problems
Survey the cutting edge of artificial intelligence research and the most important unsolved problems that define the field's future. This course covers the AGI debate, consciousness and AI, multimodal foundation models, neurosymbolic approaches, world models, AI for scientific discovery, quantum machine learning, and the environmental costs of AI. You will develop the skills to read and critically evaluate current ML research papers and understand where the field is heading.
22 lessons
- AGI Definitions & Debates
- Multimodal Foundation Model Architectures
- Limitations of Pure Neural & Pure Symbolic AI
- AI for Scientific Discovery
- Energy & Environmental Costs of AI
- AGI Timeline Predictions
- World Models
- Neurosymbolic Integration Strategies
- AI for Mathematics & Theorem Proving
- Open Research Questions in AI
- Consciousness & AI
- Vision-Language Integration
- Program Synthesis & Reasoning
- Quantum Machine Learning
- How to Read ML Research Papers
- Understanding vs Computation
- Emergent Multimodal Capabilities
- Knowledge Graphs & LLM Integration
- AI for Weather & Climate Science
- Evaluating AI Claims & Benchmarks
- Abstract Reasoning & Abstraction Learning
- Pathways to Contributing to AI Research
Unit 29
Natural Language Understanding
Explore how machines process, interpret, and generate human language. This course covers the full NLP pipeline from text preprocessing and tokenization through modern large language models, examining core tasks like sentiment analysis, named entity recognition, question answering, and summarization. You will develop a deep understanding of transformer-based architectures (BERT, GPT), prompt engineering principles, and the challenges of hallucination and grounding that define the current frontier of language AI.
25 lessons
- Text Normalization
- Language Modeling Objectives
- Question Answering Systems
- BERT Architecture & Masked Language Modeling
- Prompt Engineering Principles
- Tokenization Strategies
- Sentiment Analysis
- Text Summarization
- GPT Architecture & Autoregressive Generation
- Chain-of-Thought Reasoning
- Subword Encoding (BPE & WordPiece)
- Named Entity Recognition
- Machine Translation
- Scaling Laws & Emergent Abilities
- Hallucination & Confabulation
- Word Embeddings
- Text Classification
- Decoding Strategies
- Instruction Tuning & RLHF
- Retrieval-Augmented Generation (RAG)
- Contextual Embeddings
- Sequence Labeling
- Generation Evaluation Metrics
- In-Context Learning & Few-Shot Prompting
- Grounding & Factual Verification
Unit 30
AI for Decision Making
Examine how artificial intelligence supports and transforms decision-making across high-stakes domains. This course bridges decision theory and causal reasoning with practical AI applications in healthcare, finance, and criminal justice, while critically evaluating when AI decision support helps and when it causes harm. You will learn to apply utility functions, multi-criteria decision frameworks, counterfactual reasoning, and human-in-the-loop design principles that ensure AI augments rather than replaces sound human judgment.
23 lessons
- Decision Making Under Uncertainty
- Correlation vs Causation in AI
- Clinical Decision Support Systems
- Algorithmic Trading
- Human-AI Decision Teaming
- Utility Functions & Preferences
- Counterfactual Reasoning
- AI-Assisted Diagnosis
- AI Credit Scoring
- Automation Bias
- Multi-Criteria Decision Making
- Treatment Effect Estimation
- AI for Treatment Recommendations
- AI Fraud Detection
- Levels of AI Autonomy
- Value of Information
- Causal Graphs & DAGs
- Ethics of Healthcare AI
- Criminal Justice Risk Assessment
- When NOT to Use AI
- Heterogeneous Treatment Effects
- Predictive Policing
- Decision Audit & Accountability
Explore more roadmaps
Frequently asked questions
How long is the AI & Machine Learning roadmap?
About 1000 hours in all: 670 lessons across 30 units.
What does the AI & Machine Learning roadmap cover?
30 units, starting with Mathematical Foundations for Machine Learning, What Is Artificial Intelligence, Model Evaluation and Selection and How Machines Learn, and 26 more.
How do I start the AI & Machine Learning roadmap?
Get the Mochivia app on the App Store and open this roadmap there.