Coming soon
Artificial Intelligence & Machine Learning
Become the engineer who builds AI, not just uses it.
The complete course
- modules
- 98
- topics
- 1,118
- min lessons
- 15
What you'll be able to do.
Frame any real problem as an AI task
Stage 2 · Search, Logic & Reasoning
Train, tune and evaluate classic models
Stage 3 · Machine Learning Core
Build and debug deep neural networks
Stage 4 · Deep & Reinforcement Learning
Fine-tune and align language models
Stage 5 · Generative AI & Agents
Ship RAG pipelines and tool-using agents
Stage 5 · Generative AI & Agents
Deploy, monitor and secure AI in production
Stage 6 · Production, Trust & Research
Seven stages.One steady climb.
Each stage stands on the last — from the math under every model to a capstone of your own.
98modules
in 7 stages
1,118topics
≈ 11 per module
≈280hours
of 15-minute lessons
≈3years
≈18months
≈9months
to finish the whole course
- 1
Foundations & Math
The math under every model
7 modules · 77 topics · ≈ 19 hr
- 2
Search, Logic & Reasoning
How machines search and reason
9 modules · 100 topics · ≈ 25 hr
- 3
Machine Learning Core
Models that learn from data
21 modules · 233 topics · ≈ 58 hr
- 4
Deep & Reinforcement Learning
Networks that see, read and act
24 modules · 276 topics · ≈ 69 hr
- 5
Generative AI & Agents
LLMs, RAG and agents
18 modules · 211 topics · ≈ 53 hr
- 6
Production, Trust & Research
AI that holds up in the real world
15 modules · 178 topics · ≈ 45 hr
- 7
Practicum & Capstone
Your own system, end to end
4 modules · 43 topics · ≈ 11 hr
Months
01224048120246
MonthsFinish · ≈ 3 years · ≈ 18 months · ≈ 9 months
Hours and months are estimates: one 15-minute lesson per topic, every day.
Every module.Every topic.
Stage 1
Foundations & Math
The math under every model
7 modules · 77 topics
1The Landscape of Artificial Intelligence10 topics
- Artificial Intelligence, Machine Learning, and Deep Learning
- Symbolic, Statistical, and Neural Approaches
- Prediction, Generation, Reasoning, and Decision-Making
- Specialized and General-Purpose Models
- Learning Systems and Programmed Rules
- Capabilities, Limitations, and Evaluation Evidence
- Historical Shifts in AI Methods
- Data, Compute, and Algorithmic Progress
- Research Models and Operational AI Systems
- Boundaries Between AI Engineering and Data Analytics
2Intelligent Agents and Problem Formulation11 topics
- Agents, Environments, Observations, and Actions
- Goals, Rewards, Utilities, and Constraints
- Fully and Partially Observable Environments
- Deterministic and Stochastic Environments
- Episodic and Sequential Tasks
- Single-Agent and Multi-Agent Settings
- Reactive, Deliberative, and Learning Agents
- Translating User Needs Into AI Tasks
- Choosing Prediction and Decision Boundaries
- Defining Success, Failure, and Resource Budgets
- Establishing Baselines Before Selecting Models
3Linear Algebra for Learning Systems11 topics
- Vectors, Matrices, and Higher-Order Tensors
- Matrix Products and Batched Operations
- Norms, Distances, and Similarity Measures
- Linear Independence, Rank, and Null Spaces
- Projections and Least-Squares Geometry
- Eigenvalues and Eigenvectors in Learning
- Singular Value Decomposition
- Positive Semidefinite Matrices
- Low-Rank Approximations and Embedding Spaces
- Conditioning and Numerical Sensitivity
- Tensor Shapes and Broadcasting Rules
4Calculus and Automatic Differentiation11 topics
- Derivatives of Scalar and Vector Functions
- Gradients, Jacobians, and Hessians
- Multivariable Chain Rules
- Directional Derivatives and Local Approximations
- Matrix Calculus for Model Parameters
- Computational Graphs and Differentiable Operations
- Forward-Mode and Reverse-Mode Differentiation
- Jacobian-Vector and Vector-Jacobian Products
- Nondifferentiable Points and Subgradients
- Gradient Checking With Numerical Differences
- Differentiation Through Iterative Computations
5Probability and Information for Machine Learning11 topics
- Random Variables and Probability Distributions
- Joint, Conditional, and Marginal Probabilities
- Independence and Conditional Independence
- Bayes' Rule in Model Learning
- Expectations, Variance, and Covariance
- Likelihoods and Log-Likelihoods
- Entropy and Cross-Entropy
- Kullback-Leibler Divergence
- Mutual Information and Representation Quality
- Sampling and Monte Carlo Estimates
- Numerical Stability in Probability Calculations
6Optimization and Numerical Methods12 topics
- Objective Functions and Feasible Solutions
- Convex and Nonconvex Optimization
- First-Order and Second-Order Information
- Gradient Descent and Step-Size Selection
- Stochastic Gradients and Mini-Batches
- Constraints, Penalties, and Lagrange Multipliers
- Coordinate Descent and Proximal Methods
- Local Minima, Saddle Points, and Flat Regions
- Convergence Criteria and Optimization Diagnostics
- Floating-Point Precision and Numerical Errors
- Stable Logarithms, Exponentials, and Normalization
- Optimization Error and Modeling Error
7Computing Workflows for AI Experiments11 topics
- Array-Based and Tensor-Based Computation
- Vectorization and Batch Processing
- CPU and Accelerator Execution
- Tensor Devices, Layouts, and Data Types
- Dataset Iterators and Mini-Batch Construction
- Reproducible Randomness and Seed Management
- Experiment Configuration and Artifact Organization
- Model Checkpoints and Training State
- Profiling Runtime and Memory Use
- Numerical Reference Implementations
- Framework Abstractions and Their Tradeoffs
Stage 2
Search, Logic & Reasoning
How machines search and reason
9 modules · 100 topics
8State-Space Search and Problem Solving12 topics
- State Representations and Transition Models
- Initial States, Goals, and Path Costs
- Search Trees and State Graphs
- Breadth-First and Depth-First Strategies
- Uniform-Cost and Iterative Deepening Search
- Heuristic Estimates and Search Guidance
- A-Star Search and Heuristic Conditions
- Duplicate States and Cycle Handling
- Memory-Bounded and Anytime Search
- Search Under Partial Information
- Completeness, Optimality, and Computational Cost
- Constructing Heuristics From Problem Structure
9Local, Stochastic, and Evolutionary Search11 topics
- Optimization Through Neighboring Solutions
- Hill Climbing and Random Restarts
- Simulated Annealing
- Tabu Search and Search Memory
- Beam Search and Population-Based Search
- Genetic Representations and Variation Operators
- Selection Pressure and Population Diversity
- Evolution Strategies for Black-Box Objectives
- Multiobjective Search and Pareto Fronts
- Constraint Handling in Stochastic Search
- Comparing Search Methods Under Equal Budgets
10Constraint-Based Reasoning11 topics
- Variables, Domains, and Constraints
- Constraint Graphs and Problem Structure
- Backtracking and Variable Ordering
- Constraint Propagation and Consistency
- Global Constraints and Domain Reduction
- Boolean Satisfiability Concepts
- Satisfiability Modulo Theories
- Constraint Optimization and Soft Preferences
- Scheduling and Allocation as Constraint Problems
- Integrating Learned Heuristics With Solvers
- Explaining Infeasible or Conflicting Requirements
11Logic and Automated Inference11 topics
- Propositional and First-Order Representations
- Facts, Rules, Relations, and Quantifiers
- Syntax, Semantics, and Entailment
- Rule-Based Knowledge Systems
- Unification and Substitution
- Forward and Backward Reasoning
- Resolution and Proof Search
- Soundness and Completeness of Inference
- Default Assumptions and Nonmonotonic Reasoning
- Inconsistent Knowledge and Belief Revision
- Limits of Purely Symbolic Inference
12Knowledge Representation and Knowledge Graphs11 topics
- Entities, Relations, and Attributes
- Taxonomies, Ontologies, and Type Systems
- Events, Time, and Context in Knowledge Models
- Knowledge Graph Construction and Curation
- Entity Resolution and Relation Linking
- Schema Constraints and Knowledge Validation
- Graph Queries and Rule-Based Enrichment
- Provenance and Confidence of Knowledge Claims
- Knowledge Graph Embeddings
- Link Prediction and Graph Completion
- Combining Structured Knowledge With Learned Models
13Automated Planning11 topics
- Actions, Preconditions, and Effects
- Planning States and Goal Conditions
- Forward and Backward Planning
- Partial-Order and Hierarchical Plans
- Planning With Time and Resources
- Planning Under Uncertain Outcomes
- Conditional Plans and Belief States
- Replanning After Observed Changes
- Learned Models and Planning Heuristics
- Plan Verification and Execution Monitoring
- Planning Quality and Computational Tradeoffs
14Game Playing and Strategic Interaction11 topics
- Game States, Actions, and Payoffs
- Minimax Reasoning and Evaluation Functions
- Alpha-Beta Search and Move Ordering
- Stochastic Game Trees
- Monte Carlo Tree Search
- Imperfect Information and Hidden State
- Cooperative and Competitive Objectives
- Best Responses and Equilibrium Concepts
- Self-Play and Opponent Diversity
- Exploitability and Robust Strategy Evaluation
- Combining Search With Learned Policies and Values
15Probabilistic Graphical Models11 topics
- Graph Structure and Conditional Dependence
- Bayesian Networks and Factorization
- Markov Random Fields and Undirected Models
- Factor Graphs and Local Potentials
- Hidden Variables and Structured Uncertainty
- Hidden Markov Models
- Dynamic Bayesian Networks
- Conditional Random Fields
- Learning Graph Parameters and Structure
- Modeling Missing and Partially Observed Variables
- Choosing Graphical Models for Structured Problems
16Probabilistic Inference Methods11 topics
- Exact Marginalization and Variable Elimination
- Message Passing and Belief Propagation
- Inference Complexity and Graph Structure
- Maximum A Posteriori Inference
- Importance Sampling and Weight Degeneracy
- Markov Chain Monte Carlo Concepts
- Gibbs and Metropolis-Hastings Sampling
- Sequential Monte Carlo and Particle Methods
- Variational Approximations
- Evidence Lower Bounds and Optimization
- Diagnosing Approximate Inference Quality
Stage 3
Machine Learning Core
Models that learn from data
21 modules · 233 topics
17Learning Paradigms and Objective Design11 topics
- Supervised and Unsupervised Learning
- Self-Supervised and Reinforcement Learning
- Parametric and Nonparametric Models
- Discriminative and Generative Learning
- Batch, Online, and Incremental Learning
- Empirical Risk Minimization
- Loss Functions and Task Alignment
- Regularization as an Inductive Preference
- Point Predictions and Predictive Distributions
- Multiobjective and Cost-Sensitive Learning
- Matching Learning Paradigms to Available Feedback
18Training Data and Annotation12 topics
- Defining the Target Population and Unit of Prediction
- Data Provenance and Permitted Use
- Label Definitions and Annotation Guidelines
- Human Annotation and Agreement Analysis
- Label Noise and Ambiguous Examples
- Class Balance and Coverage of Rare Cases
- Duplicate and Near-Duplicate Examples
- Training Data Contamination
- Dataset Versioning and Documentation
- Synthetic Data and Its Limitations
- Data Quality Improvements as Model Interventions
- Dataset Growth and Collection Priorities
19Features and Model Input Representations12 topics
- Numerical and Categorical Feature Representations
- Scaling, Normalization, and Model Sensitivity
- Missing Values and Missingness Indicators
- Category Encoding and High-Cardinality Inputs
- Feature Interactions and Basis Expansions
- Sparse and Dense Representations
- Learned Embeddings for Discrete Inputs
- Feature Selection and Redundant Inputs
- Preprocessing Fit Boundaries
- Temporal Availability of Features
- Feature Leakage and Proxy Information
- Reproducible Model Input Pipelines
20Training, Validation, and Generalization Design11 topics
- Training, Validation, and Test Roles
- Random, Stratified, and Grouped Splits
- Time-Aware and Forward Evaluation
- Cross-Validation and Nested Selection
- Leakage Across Related Examples
- Distribution Differences Between Development and Use
- Validation Set Reuse and Adaptive Overfitting
- Learning Curves and Data Requirements
- Baseline Comparisons and Ablation Studies
- Uncertainty Across Data Splits and Random Seeds
- Holding Out Realistic Failure Scenarios
21Learning Theory and Inductive Bias11 topics
- Hypothesis Classes and Model Capacity
- Bias, Variance, and Approximation Error
- Generalization Error and Expected Risk
- Overfitting and Underfitting Mechanisms
- Sample Complexity and PAC Learning Concepts
- VC Dimension and Complexity Measures
- Regularization and Stability Connections
- No-Free-Lunch Results and Their Interpretation
- Interpolation and Double Descent
- Implicit Bias of Optimization
- Limits of Theory in Modern Deep Learning
22Predictive Model Evaluation12 topics
- Regression Error and Scale Sensitivity
- Classification Confusion Patterns
- Precision, Recall, and Class-Specific Tradeoffs
- ROC and Precision-Recall Curves
- Threshold Selection Under Unequal Costs
- Log Loss and Proper Scoring Rules
- Probability Calibration and Reliability Diagrams
- Macro, Micro, and Weighted Aggregation
- Multilabel and Structured Prediction Metrics
- Subgroup and Slice-Based Error Analysis
- Abstention and Selective Prediction
- Matching Evaluation Metrics to Deployment Decisions
23Model Selection and Hyperparameter Optimization11 topics
- Hyperparameters and Learned Parameters
- Grid, Random, and Adaptive Search
- Bayesian Optimization for Model Selection
- Early-Stopping and Resource Allocation Strategies
- Successive Halving and Multi-Fidelity Evaluation
- Conditional and Structured Search Spaces
- Search Budget and Fair Model Comparisons
- Repeated Trials and Ranking Uncertainty
- Tracking Experiments and Model Lineage
- Selecting Models Under Latency and Memory Constraints
- Final Evaluation After Model Selection
24Linear Predictors and Regularized Regression11 topics
- Linear Prediction as a Learning Model
- Squared, Absolute, and Robust Regression Losses
- Closed-Form and Iterative Parameter Estimation
- Polynomial and Basis-Function Models
- Ridge, Lasso, and Elastic Net Penalties
- Sparse Coefficients and Correlated Features
- Multioutput Linear Prediction
- Quantile Regression Objectives
- Online Updates for Linear Models
- Optimization and Scaling Considerations
- Linear Models as Diagnostic Baselines
25Probabilistic Classification Models11 topics
- Binary Logistic Prediction
- Multiclass Softmax Models
- Decision Boundaries and Class Probabilities
- Cross-Entropy Training Objectives
- Regularized Classification
- Naive Bayes Model Families
- Gaussian Discriminant Models
- Generative and Discriminative Classifiers
- Class Weighting and Prior Probability Changes
- Multilabel Classification Strategies
- Calibrating and Comparing Simple Classifiers
26Instance-Based and Distance-Based Learning10 topics
- Nearest-Neighbor Classification and Regression
- Distance Functions and Feature Scaling
- Neighborhood Size and Local Smoothing
- Distance-Weighted Predictions
- High-Dimensional Geometry and Distance Concentration
- Prototype and Centroid-Based Models
- Locally Weighted Regression
- Learned Distance Metrics
- Search Cost and Approximate Neighbor Retrieval
- Memory Requirements and Incremental Updates
27Kernel Methods and Support Vector Machines11 topics
- Feature Maps and Kernel Similarity
- Positive Semidefinite Kernel Functions
- Maximum-Margin Classification
- Hard and Soft Margin Objectives
- Support Vectors and Decision Functions
- Dual Optimization and Kernelization
- Linear, Polynomial, and Radial Basis Kernels
- Support Vector Regression
- Kernel Hyperparameters and Scaling
- Kernel Approximation for Larger Datasets
- Computational Limits of Kernel Models
28Decision Trees and Rule Learning10 topics
- Recursive Partitioning of Feature Space
- Impurity and Information-Based Split Criteria
- Regression Tree Objectives
- Categorical and Missing-Value Handling
- Tree Depth and Minimum Leaf Constraints
- Pruning and Complexity Control
- Instability of Individual Trees
- Rule Extraction and Decision Paths
- Monotonicity and Domain Constraints
- Interpreting Tree Predictions and Failure Regions
29Ensemble Learning and Boosting12 topics
- Diversity and Error Reduction in Ensembles
- Bootstrap Aggregation
- Random Forest Construction
- Randomized Tree Ensembles
- Adaptive Boosting and Reweighted Examples
- Gradient Boosting as Functional Optimization
- Second-Order Boosting Objectives
- Shrinkage, Subsampling, and Early Stopping
- Histogram-Based and Categorical Boosting Techniques
- Stacking and Out-of-Fold Predictions
- Voting and Probability Averaging
- Ensemble Cost, Calibration, and Interpretability
30Bayesian Learning and Gaussian Processes11 topics
- Priors, Likelihoods, and Posterior Learning
- Bayesian Linear Prediction
- Posterior Predictive Distributions
- Model Evidence and Complexity Preferences
- Gaussian Processes as Distributions Over Functions
- Covariance Functions and Kernel Design
- Gaussian Process Regression
- Gaussian Process Classification Concepts
- Hyperparameter Learning Through Marginal Likelihood
- Sparse Approximations and Inducing Variables
- Uncertainty Quality Under Model Misspecification
31Clustering Algorithms11 topics
- Similarity-Based Grouping and Cluster Assumptions
- K-Means and Centroid Optimization
- Initialization and Cluster Stability
- Hierarchical and Agglomerative Clustering
- Density-Based Cluster Discovery
- Spectral Clustering and Graph Structure
- Soft Assignments and Overlapping Groups
- Constraints and Prior Knowledge in Clustering
- Internal and External Cluster Evaluation
- High-Dimensional and Unequal-Density Challenges
- Interpreting Clusters Without Inventing Categories
32Dimensionality Reduction and Latent Structure11 topics
- Linear Projection and Variance Preservation
- Principal Component Analysis Algorithms
- Truncated SVD for Sparse Inputs
- Independent Component Analysis
- Nonnegative Matrix Factorization
- Random Projections and Distance Preservation
- Manifold Learning and Neighborhood Structure
- Nonlinear Embedding Methods
- Reconstruction, Compression, and Predictive Utility
- Visualization Distortion and Interpretation Limits
- Evaluating Reduced Representations Downstream
33Mixture Models and Density Estimation11 topics
- Latent Components and Mixture Distributions
- Gaussian Mixture Models
- Expectation-Maximization Updates
- Initialization and Local Optima in EM
- Component Number and Model Selection
- Covariance Constraints and Singular Solutions
- Kernel Density Estimation
- Conditional Density Models
- Missing Data Through Latent Variable Models
- Density Estimation in High Dimensions
- Likelihood Quality and Task Usefulness
34Anomaly and Novelty Detection11 topics
- Point, Contextual, and Collective Anomalies
- Novelty Detection and Out-of-Distribution Inputs
- Distance and Local Density Methods
- Isolation-Based Models
- One-Class Classification
- Reconstruction-Based Anomaly Scores
- Sequence and Context-Aware Detection
- Rare Positive Labels and Contaminated Training Sets
- Thresholds Under False Alarm Constraints
- Evaluation Under Extreme Class Imbalance
- Analyst Feedback and Changing Definitions of Normality
35Semi-Supervised and Weakly Supervised Learning11 topics
- Learning From Labeled and Unlabeled Examples
- Assumptions Behind Semi-Supervised Learning
- Pseudo-Labels and Confidence Thresholds
- Consistency-Based Training
- Graph-Based Label Propagation
- Multi-View and Co-Training Concepts
- Learning From Noisy Labeling Rules
- Label Model Conflicts and Correlations
- Positive-Unlabeled Learning
- Multiple-Instance Learning
- Evaluating Gains Against Labeling Effort
36Active Learning and Human Feedback11 topics
- Pool-Based and Stream-Based Querying
- Uncertainty Sampling
- Diversity and Coverage in Acquisition
- Query-by-Committee Methods
- Batch Selection and Annotation Efficiency
- Expected Improvement and Value of Information
- Labeling Cost and Annotator Expertise
- Human Review of Ambiguous Examples
- Feedback Loops and Selection Bias
- Stopping Criteria for Data Acquisition
- Comparing Active Learning With Random Sampling
37Transfer, Multitask, and Domain Adaptation11 topics
- Source and Target Learning Problems
- Reusing Features and Learned Representations
- Shared Backbones and Task-Specific Heads
- Task Balancing and Gradient Interference
- Negative Transfer and Task Compatibility
- Covariate and Label Distribution Shifts
- Domain-Invariant Representations
- Importance Weighting and Adaptation Assumptions
- Few-Shot and Meta-Learning Concepts
- Domain Generalization Without Target Labels
- Measuring Transfer Benefits and Failure Cases
Stage 4
Deep & Reinforcement Learning
Networks that see, read and act
24 modules · 276 topics
38Neural Network Foundations11 topics
- Artificial Neurons and Weighted Computation
- Multilayer Perceptron Architectures
- Activation Functions and Their Effects
- Hidden Representations and Feature Composition
- Output Layers for Different Learning Tasks
- Parameter Sharing and Architectural Bias
- Depth, Width, and Expressive Capacity
- Residual Connections and Information Flow
- Initialization and Signal Propagation
- Neural Networks as Function Approximators
- Architectural Choices Under Resource Constraints
39Backpropagation and Training Implementation11 topics
- Forward Passes and Loss Computation
- Backpropagation Through Layered Computation
- Parameter Gradients and Update Steps
- Automatic Differentiation in Neural Frameworks
- Implementing Neural Models With PyTorch
- Gradient Accumulation and Reset Behavior
- Training and Evaluation Modes
- Detaching Computations and Freezing Parameters
- Custom Losses and Differentiable Components
- Checkpointing Model and Optimizer State
- Verifying a Training Loop on Small Examples
40Optimization for Deep Networks12 topics
- Stochastic Gradient Descent With Momentum
- Adaptive Gradient Methods
- Adam and Decoupled Weight Decay
- Learning Rate Schedules and Warmup
- Batch Size and Gradient Noise
- Gradient Clipping and Exploding Updates
- Mixed-Precision Optimization
- Loss Scaling and Numerical Overflow
- Gradient Accumulation for Large Effective Batches
- Optimizer State and Memory Requirements
- Diagnosing Optimization Instability
- Relating Training Speed to Final Model Quality
41Regularization and Deep Generalization12 topics
- Weight Penalties and Parameter Constraints
- Dropout and Stochastic Computation
- Early Stopping and Checkpoint Selection
- Data Augmentation and Invariance Assumptions
- Label Smoothing
- Mixup and Interpolated Training Examples
- Batch Normalization and Layer Normalization
- Normalization Effects on Optimization and Generalization
- Representation Bottlenecks
- Sharpness, Robustness, and Generalization Hypotheses
- Overparameterization and Memorization
- Separating Regularization Benefits Through Ablations
42Neural Network Debugging12 topics
- Testing Data and Labels Before Training
- Overfitting a Small Batch as a Diagnostic
- Inspecting Activations and Gradient Distributions
- Detecting Dead Units and Saturation
- Diagnosing Vanishing and Exploding Gradients
- Tracking Loss Components and Scale Mismatches
- Identifying NaNs and Numerical Instability
- Finding Train-Evaluation Behavior Differences
- Detecting Incorrect Masking and Padding
- Checking Data Pipeline and Device Bottlenecks
- Structured Ablations for Model Failures
- Building Minimal Reproducible Training Experiments
43Convolutional Neural Networks11 topics
- Convolution as a Learned Local Operator
- Filters, Channels, Stride, and Padding
- Receptive Fields and Spatial Resolution
- Pooling and Downsampling
- Translation Equivariance and Its Limits
- Residual Convolutional Architectures
- Depthwise and Grouped Convolutions
- Dilated Convolutions and Context Range
- Encoder-Decoder Feature Hierarchies
- Pretrained Visual Backbones
- Convolutional Model Efficiency and Scaling
44Learning for Visual Recognition12 topics
- Image Classification and Multilabel Recognition
- Object Detection Formulations
- Localization Losses and Bounding Box Predictions
- Semantic and Instance Segmentation
- Keypoint and Pose Prediction Objectives
- Vision Transformers and Patch Representations
- Dense Prediction and Multiscale Features
- Imbalanced Visual Labels and Small Objects
- Visual Dataset Bias and Shortcut Learning
- Task-Specific Visual Evaluation Metrics
- Transfer Learning Across Visual Domains
- Boundaries Between Model Learning and Vision System Engineering
45Recurrent and Sequential Neural Models11 topics
- Sequence Inputs and Hidden States
- Recurrent Neural Network Computation
- Backpropagation Through Time
- Long Short-Term Memory Networks
- Gated Recurrent Units
- Bidirectional Sequence Encoders
- Sequence-to-Sequence Architectures
- Teacher Forcing and Exposure Bias
- Variable-Length Sequences and Masking
- Temporal Convolutional Alternatives
- Long-Range Dependencies and Memory Limits
46Attention and Transformer Architectures12 topics
- Query, Key, and Value Representations
- Scaled Dot-Product Attention
- Self-Attention and Cross-Attention
- Multihead Attention
- Causal and Padding Masks
- Positional Embeddings and Rotary Position Methods
- Transformer Feedforward Blocks
- Residual Paths and Normalization Placement
- Encoder, Decoder, and Encoder-Decoder Models
- Attention Complexity and Context Length
- Training Objectives for Transformer Families
- Diagnosing Attention and Position Handling Errors
47Efficient Sequence Models and Sparse Architectures11 topics
- Sparse and Local Attention Patterns
- Linear Attention Approximations
- Recurrent Computation in Sequence Models
- State-Space Sequence Modeling
- Selective State Updates
- Hybrid Attention and State-Space Architectures
- Mixture-of-Experts Layers
- Expert Routing and Load Balancing
- Active Parameters and Total Model Capacity
- Context Length and Memory Tradeoffs
- Comparing Architectures Under Matched Compute Budgets
48Graph and Geometric Learning12 topics
- Node, Edge, and Graph Prediction Tasks
- Message Passing and Neighborhood Aggregation
- Graph Convolution and Graph Attention
- Graph-Level Pooling and Readout
- Transductive and Inductive Graph Learning
- Heterogeneous and Temporal Graphs
- Oversmoothing and Oversquashing
- Graph Sampling and Scalability
- Permutation Invariance and Equivariance
- Geometric Symmetries in Learned Models
- Point Sets and Spatially Structured Inputs
- Graph Evaluation Splits and Relational Leakage
49Self-Supervised and Metric Learning11 topics
- Constructing Supervision From Unlabeled Data
- Positive Pairs, Negative Pairs, and Augmentations
- Contrastive Learning Objectives
- Triplet and Ranking-Based Metric Losses
- Negative Sampling and Batch Composition
- Masked Input Reconstruction
- Teacher-Student and Self-Distillation Methods
- Representation Collapse and Prevention Mechanisms
- Embedding Geometry and Transfer Quality
- Linear Probes and Fine-Tuning Evaluation
- Choosing Self-Supervised Tasks for a Data Domain
50Text Representations and Tokenization11 topics
- Text Normalization and Language Variation
- Word, Subword, Byte, and Character Representations
- Byte-Pair and Unigram Tokenization
- Vocabulary Size and Sequence Length Tradeoffs
- Tokenization of Multilingual and Technical Text
- Special Tokens and Conversation Formatting
- Sparse Text Features and Classical Baselines
- Static and Contextual Word Embeddings
- Sentence and Document Representations
- Tokenization Errors and Lost Information
- Tokenizer Consistency Across Training and Inference
51Natural Language Understanding Tasks11 topics
- Text and Document Classification
- Named Entity and Relation Extraction
- Sequence Labeling and Span Prediction
- Semantic Similarity and Paraphrase Detection
- Natural Language Inference
- Extractive Question Answering
- Semantic Parsing and Structured Meaning
- Summarization and Information Preservation
- Translation and Cross-Lingual Transfer
- Ambiguity, Pragmatics, and Context Dependence
- Evaluating Language Tasks Beyond Surface Matching
52Speech and Audio Learning11 topics
- Waveforms, Spectrograms, and Acoustic Features
- Audio Encoders and Temporal Resolution
- Speech Recognition Learning Objectives
- Connectionist Temporal Classification
- Transducer and Encoder-Decoder Speech Models
- Speaker Representations and Diarization
- Audio Event Classification and Detection
- Self-Supervised Speech Representations
- Speech Synthesis and Neural Vocoders
- Noise, Accents, and Domain Variation
- Audio Quality, Intelligibility, and Task Evaluation
53Multimodal Representation Learning12 topics
- Aligned and Unaligned Modalities
- Early, Late, and Intermediate Fusion
- Shared Embedding Spaces Across Modalities
- Image-Text Contrastive Learning
- Cross-Modal Attention and Conditioning
- Vision-Language Understanding
- Video and Temporal Multimodal Inputs
- Audio-Visual Learning
- Missing and Conflicting Modalities
- Grounding Predictions in Input Evidence
- Multimodal Dataset Alignment and Bias
- Evaluating Cross-Modal Transfer and Robustness
54Recommendation, Matching, and Learning to Rank12 topics
- User, Item, and Context Representations
- Collaborative Filtering and Matrix Factorization
- Implicit Feedback and Negative Sampling
- Two-Tower Retrieval Models
- Pointwise, Pairwise, and Listwise Ranking
- Neural Recommendation Architectures
- Sequential Recommendation
- Cold-Start and Sparse-Interaction Problems
- Exposure Bias and Feedback Loops
- Diversity, Coverage, and Relevance Tradeoffs
- Offline Ranking Evaluation and Its Limits
- Retrieval and Ranking as Separate Learning Stages
55Learning From Temporal and Event Data11 topics
- Temporal Prediction Task Definitions
- Lagged Inputs and Sequence Windows
- Global Models Across Multiple Time Series
- Recurrent, Convolutional, and Attention-Based Forecasting
- Multistep Prediction and Error Accumulation
- Probabilistic Forecasts and Quantile Objectives
- Irregular Sampling and Missing Observations
- Event Sequences and Learned Intensity Models
- Covariate Availability at Prediction Time
- Temporal Distribution Shift and Retraining
- Backtesting Learned Models Without Future Leakage
56Bandits and Sequential Experimentation11 topics
- Exploration and Exploitation
- Stochastic Multi-Armed Bandits
- Regret and Sample Efficiency
- Epsilon-Greedy and Confidence-Based Strategies
- Thompson Sampling
- Contextual Bandits
- Delayed and Partial Feedback
- Logged Policies and Propensity Information
- Off-Policy Evaluation for Bandit Decisions
- Nonstationary Rewards and Adaptation
- Constraints on Online Exploration
57Reinforcement Learning Foundations12 topics
- States, Actions, Rewards, and Transitions
- Markov Decision Processes
- Policies and Trajectory Distributions
- Returns, Discounting, and Episode Boundaries
- State-Value and Action-Value Functions
- Bellman Relationships
- Policy Evaluation and Policy Improvement
- Value Iteration and Policy Iteration
- Partial Observability and Belief-State Policies
- Reward Design and Specification Errors
- Environment Design and Evaluation Protocols
- Learning Performance and Sample Efficiency
58Value-Based Reinforcement Learning12 topics
- Monte Carlo Return Estimation
- Temporal-Difference Learning
- SARSA and Q-Learning
- On-Policy and Off-Policy Updates
- Function Approximation for Value Estimates
- Deep Q-Networks
- Replay Buffers and Target Networks
- Double Estimation and Overestimation Bias
- Prioritized Replay and Sampling Effects
- Distributional Value Learning
- Exploration Strategies for Sparse Rewards
- Stability and Reproducibility in Value Learning
59Policy Gradients and Actor-Critic Methods12 topics
- Parameterized Stochastic Policies
- Policy Gradient Estimation
- REINFORCE and Variance Reduction
- Baselines and Advantage Estimates
- Actor-Critic Learning
- Generalized Advantage Estimation
- Trust Regions and Clipped Policy Objectives
- Proximal Policy Optimization
- Deterministic Policy Gradients
- Entropy-Regularized Learning and Soft Actor-Critic
- Continuous Action Constraints
- Policy Evaluation Across Seeds and Environments
60Model-Based and Hierarchical Reinforcement Learning12 topics
- Learning Transition and Reward Models
- Planning With Learned Dynamics
- Model Predictive Decision-Making
- World Models and Latent Rollouts
- Compounding Model Error
- Uncertainty-Aware Planning
- Model-Based and Model-Free Hybrid Methods
- Temporal Abstraction and Options
- Hierarchical Policies and Subgoals
- Goal-Conditioned Learning
- Simulation-to-Environment Transfer
- Evaluating Learned Models Through Decision Quality
61Offline Learning, Imitation, and Demonstrations11 topics
- Learning Policies From Fixed Datasets
- Dataset Coverage and Action Support
- Distribution Shift in Offline Policy Learning
- Conservative Value and Policy Objectives
- Offline Policy Evaluation
- Behavioral Cloning
- Interactive Imitation and Data Aggregation
- Inverse Reinforcement Learning Concepts
- Learning Rewards From Demonstrations
- Demonstration Quality and Conflicting Behavior
- Combining Offline Training With Constrained Online Updates
Stage 5
Generative AI & Agents
LLMs, RAG and agents
18 modules · 211 topics
62Generative Modeling and Variational Autoencoders11 topics
- Modeling Data Distributions and Conditional Generation
- Likelihood-Based and Implicit Generative Models
- Autoregressive Factorization Across Data Types
- Latent Variables and Learned Representations
- Deterministic Autoencoders and Reconstruction
- Variational Autoencoder Objectives
- Reparameterization and Stochastic Gradients
- Posterior Collapse and Latent Variable Use
- Conditional and Hierarchical Latent Models
- Discrete Latents and Vector Quantization
- Sample Quality, Coverage, and Likelihood Tradeoffs
63Generative Adversarial Networks10 topics
- Generator and Discriminator Objectives
- Adversarial Training as a Coupled Game
- Conditional Adversarial Generation
- Mode Collapse and Missing Diversity
- Gradient Behavior and Training Instability
- Wasserstein-Based Objectives
- Gradient Penalties and Spectral Constraints
- Image Translation and Paired Supervision
- Latent Space Manipulation
- Evaluating Adversarial Generative Models
64Normalizing Flows10 topics
- Invertible Transformations of Probability Densities
- Change-of-Variables Likelihoods
- Jacobian Determinants and Tractable Architectures
- Coupling and Autoregressive Flow Layers
- Expressive Transformations and Computational Cost
- Conditional Density Estimation With Flows
- Continuous Normalizing Flows
- Neural Differential Equations and Density Evolution
- Sampling Speed and Density Evaluation Tradeoffs
- Comparing Flow Models With Other Generative Families
65Diffusion Models and Flow Matching14 topics
- Forward Noising and Learned Reverse Processes
- Noise, Data, and Score Prediction
- Noise Schedules and Training Weighting
- Denoising Objectives and Score Matching
- Stochastic and Deterministic Sampling
- Sampler Accuracy and Generation Cost
- Latent-Space Diffusion
- Conditional Generation and Guidance
- Probability Paths and Learned Vector Fields
- Conditional Flow-Matching Objectives
- Rectified Flow Concepts
- Connections and Differences Between Generative Formulations
- Distillation Into Faster Generators
- Evaluating Quality, Diversity, and Controllability
66Image, Video, and Audio Generation11 topics
- Text-Conditioned Visual Generation
- Image Editing and Inpainting Objectives
- Spatial Conditioning and Reference Inputs
- Identity, Style, and Composition Consistency
- Video Generation and Temporal Coherence
- Motion Representations and Long-Horizon Consistency
- Audio and Music Generation Models
- Speech Generation and Speaker Conditioning
- Multimodal Generation Interfaces
- Artifact Detection and Perceptual Evaluation
- Provenance, Consent, and Permitted Media Use
67Foundation Language Model Pretraining12 topics
- Next-Token and Denoising Language Objectives
- Decoder-Only Language Model Construction
- Context Windows and Training Sequences
- Token Embeddings and Output Projections
- Vocabulary and Parameter Sharing
- Training From Scratch and Continued Pretraining
- Pretraining Loss and Downstream Capabilities
- Multilingual and Domain-Specific Language Modeling
- Sequence Packing and Attention Boundaries
- Training Stability and Checkpoint Evaluation
- Base Models and Adapted Models
- Limits of Language Modeling as a Training Objective
68Language Model Data Engineering12 topics
- Corpus Selection and Source Mixtures
- Text Extraction and Document Boundaries
- Language Identification and Quality Filtering
- Exact and Approximate Deduplication
- Benchmark and Evaluation Data Decontamination
- Personal Information and Sensitive Content Handling
- Licensed, Public, and Synthetic Training Sources
- Data Mixture Weighting and Sampling Schedules
- Synthetic Data Verification and Diversity
- Repeated Data and Memorization Risk
- Tokenization Quality Across the Corpus
- Data Ablations and Corpus Documentation
69Scaling Laws and Training Resource Planning11 topics
- Model Size, Dataset Size, and Training Compute
- Empirical Scaling Relationships
- Compute-Optimal Allocation Concepts
- Parameter Count and Active Parameter Count
- Training FLOPs and Resource Accounting
- Data Quality and Effective Dataset Size
- Context Length and Training Cost
- Small-Scale Experiments for Larger Training Decisions
- Scaling Predictions and Extrapolation Limits
- Checkpoint Frequency and Recovery Cost
- Comparing Training Plans Under Fixed Budgets
70Instruction Tuning and Efficient Adaptation12 topics
- Instruction-Response Training Data
- Supervised Fine-Tuning Objectives
- Conversation Templates and Loss Masking
- Full-Parameter Fine-Tuning
- Adapters and Low-Rank Adaptation
- Quantized Base Models and Adapter Training
- Domain Adaptation and Continued Instruction Training
- Data Mixtures and Capability Retention
- Overfitting and Forgetting During Adaptation
- Adapter Composition and Deployment Constraints
- Evaluating Adaptation Against Prompting Baselines
- Versioning Adapted Models and Training Data
71Preference Learning and Post-Training12 topics
- Preference Data and Comparison Tasks
- Reward Model Training
- Human and Model-Generated Feedback
- Reinforcement Learning From Preference Signals
- Reference Policies and Divergence Constraints
- Direct Preference Optimization
- Verifiable Rewards and Outcome Feedback
- Process Feedback and Intermediate Evaluation
- Reward Overoptimization and Proxy Failures
- Preference Diversity and Annotation Bias
- Capability Retention During Post-Training
- Evaluating Helpfulness, Reliability, and Behavioral Boundaries
72Language Model Inference and Structured Generation12 topics
- Autoregressive Decoding and Token Probabilities
- Greedy, Beam, and Sampling-Based Decoding
- Temperature and Probability Truncation
- Length, Repetition, and Stop Conditions
- Prefill and Token-by-Token Generation
- Key-Value Caching
- Streaming and Partial Outputs
- Grammar-Constrained and Schema-Constrained Generation
- Structured Output Validation
- Tool Call Representations
- Decoding Reproducibility and Nondeterminism
- Latency, Cost, and Output Quality Tradeoffs
73Prompting and Context Engineering12 topics
- Task Instructions and Success Criteria
- Zero-Shot and Few-Shot Examples
- Demonstration Selection and Ordering
- Separating Instructions From Reference Material
- Context Budgets and Information Prioritization
- Document Selection and Context Assembly
- Conversation State and Summarized History
- Structured Inputs and Output Contracts
- Asking for Evidence and Verifiable Results
- Prompt Sensitivity and Controlled Experiments
- Context Failure Modes and Conflicting Instructions
- Versioning and Evaluating Prompt Templates
74Neural Retrieval and Semantic Search12 topics
- Query and Document Representation Learning
- Bi-Encoders and Cross-Encoders
- Contrastive Retrieval Training
- In-Batch and Hard Negative Sampling
- Dense, Sparse, and Hybrid Retrieval
- Approximate Nearest-Neighbor Index Interfaces
- Chunk and Document-Level Retrieval
- Candidate Generation and Reranking
- Query Reformulation and Expansion
- Retrieval Recall and Ranking Quality
- Domain Adaptation of Embedding Models
- Index Freshness and Representation Version Changes
75Retrieval-Augmented Generation12 topics
- Retrieval, Context Assembly, and Grounded Generation
- Document Parsing and Chunking Strategies
- Metadata Filters and Access-Aware Retrieval
- Multi-Stage Retrieval Pipelines
- Query Decomposition and Multi-Hop Retrieval
- Grounding Answers in Retrieved Evidence
- Citations and Source Attribution Quality
- Missing, Stale, and Contradictory Evidence
- Abstention When Retrieval Is Insufficient
- Evaluating Retrieval and Generation Separately
- Comparing RAG, Long Context, and Fine-Tuning
- End-to-End RAG Error Analysis
76Reasoning and Inference-Time Computation12 topics
- Task Decomposition and Intermediate Representations
- Search Over Candidate Solutions
- Multiple Samples and Self-Consistency
- Correlated Errors Across Generated Solutions
- Verifiers and Outcome-Based Selection
- External Calculators, Solvers, and Execution Feedback
- Reasoning With Structured Constraints
- Inference-Time Compute Allocation
- Learned Reasoning and Generalization Limits
- Reported Rationales and Faithful Explanations
- Evaluating Reasoning on Verifiable Tasks
- Benchmark Contamination and Apparent Reasoning Gains
77Tool-Using AI Agents12 topics
- Agent Loops and Environment Interfaces
- Tool Definitions and Argument Schemas
- Planning, Acting, Observing, and Revising
- Tool Selection and Action Validation
- Permissions and Scoped Capabilities
- Instruction Authority and Untrusted Tool Results
- Timeouts, Retries, and Idempotent Actions
- Sandboxed Code Execution Interfaces
- Human Approval for Consequential Actions
- Termination Conditions and Resource Budgets
- Recovery From Tool Errors
- Recording Agent Actions and Evidence
78Agent Memory and Multi-Agent Systems12 topics
- Working Context and Persistent Memory
- Memory Selection, Summarization, and Retrieval
- Memory Freshness and Contradiction Handling
- Separating User Facts From Model Inferences
- Workflow Graphs and Agent Coordination
- Specialized Roles and Task Delegation
- Shared State and Communication Protocols
- Multi-Agent Disagreement and Resolution
- Cascading Errors and Shared Blind Spots
- Centralized and Distributed Agent Supervision
- Comparing Multi-Agent Designs With Simpler Workflows
- Coordinated Resource and Permission Limits
79Agent Evaluation and Reliability12 topics
- Task Success and Environment-State Verification
- Tool Choice and Argument Accuracy
- Trajectory-Level Evaluation
- Simulated Environments and Replayable Tasks
- Partial Progress and Recovery Metrics
- Long-Horizon Error Accumulation
- User Intervention and Supervision Requirements
- Cost, Latency, and Reliability Budgets
- Adversarial and Unexpected Tool Responses
- Evaluating Memory and Delegation Failures
- Regression Suites for Agent Behavior
- Deployment Boundaries From Evaluation Evidence
Stage 6
Production, Trust & Research
AI that holds up in the real world
15 modules · 178 topics
80Online and Continual Learning11 topics
- Streaming Data and Incremental Updates
- Stationary and Changing Learning Environments
- Catastrophic Forgetting
- Experience Replay and Memory Selection
- Regularization-Based Continual Learning
- Expanding and Modular Model Architectures
- Task, Domain, and Class Incremental Settings
- Measuring Forgetting and Forward Transfer
- Update Frequency and Stability Tradeoffs
- Evaluation Without Access to Future Data
- Controlled Model Updates and Rollback Criteria
81Federated and Privacy-Preserving Learning12 topics
- Centralized and Federated Training Assumptions
- Client Sampling and Local Optimization
- Federated Averaging and Aggregation Variants
- Nonidentical Client Data Distributions
- Communication and Computation Tradeoffs
- Personalization Across Clients
- Secure Aggregation Interfaces
- Differential Privacy and Privacy Budgets
- Clipping, Noise, and Utility Tradeoffs
- Privacy Accounting Across Training Steps
- Evaluating Leakage and Privacy Limitations
- Robustness to Unreliable Client Updates
82Causal and Invariant Learning11 topics
- Predictive Association and Causal Structure
- Structural Causal Models for Learning Problems
- Interventions and Counterfactual Queries
- Causal Discovery Assumptions
- Confounding and Selection in Training Data
- Invariant Prediction Across Environments
- Causal Representation Learning Concepts
- Counterfactual Data Augmentation
- Causal Models for Decision Policies
- Testing Transportability and Domain Assumptions
- Limits of Causal Claims From Observational Learning
83Neurosymbolic and Constraint-Guided Learning11 topics
- Combining Neural Representations With Symbolic Structure
- Learned Perception and Logical Reasoning Interfaces
- Rules as Features, Constraints, and Loss Terms
- Differentiable Reasoning Components
- Neural Program Synthesis Concepts
- Learning to Guide Symbolic Search
- Knowledge Graphs in Neural Predictions
- Constraint Satisfaction for Generated Outputs
- Verifiable Intermediate Representations
- Tradeoffs Between Flexibility and Formal Structure
- Evaluating Generalization Beyond Training Templates
84Scientific and Physics-Informed Machine Learning12 topics
- Scientific Data and Domain Constraints
- Surrogate Models for Expensive Simulations
- Physics-Informed Training Objectives
- Boundary and Initial Condition Constraints
- Neural Operators and Function-to-Function Learning
- Symmetry-Aware Scientific Representations
- Learning Dynamical Systems
- Inverse Problems and Parameter Identification
- Multi-Fidelity and Simulation-Based Training
- Physical Validity and Conservation Checks
- Uncertainty in Scientific Model Predictions
- Evaluating Extrapolation Beyond Observed Conditions
85Model Compression and Edge Intelligence12 topics
- Model Size, Compute, and Memory Footprint
- Weight and Activation Quantization
- Post-Training and Quantization-Aware Methods
- Structured and Unstructured Pruning
- Knowledge Distillation Objectives
- Low-Rank and Factorized Model Approximations
- Sparse Computation and Hardware Realities
- Compact Architectures for On-Device Inference
- Accuracy, Latency, Energy, and Privacy Tradeoffs
- Calibration Data for Compression
- Verifying Compressed Models Across Input Slices
- Compression Effects on Rare and Sensitive Behaviors
86Distributed and Efficient Model Training12 topics
- Training Memory for Parameters, Gradients, and Optimizers
- Data Parallelism and Gradient Synchronization
- Sharded Parameters and Optimizer States
- Tensor and Pipeline Parallelism
- Expert Parallelism in Sparse Models
- Communication Overhead and Scaling Efficiency
- Activation Checkpointing and Recomputation
- Mixed Precision and Numerical Consistency
- Efficient Attention and Memory Access
- Input Pipeline and Accelerator Utilization
- Distributed Checkpoints and Failure Recovery
- Correctness Checks for Parallel Training
87Model Inference and Serving12 topics
- Batch, Online, and Streaming Prediction
- Model Input and Output Contracts
- Export Formats and Runtime Compatibility
- Dynamic Batching and Request Scheduling
- Throughput and Tail Latency
- Model and Result Caching
- Language Model Cache Management
- Speculative Decoding and Verification
- Routing Between Models of Different Capabilities
- Resource Limits and Graceful Degradation
- Numerical Parity Across Inference Runtimes
- Quality-Aware Inference Optimization
88Machine Learning Lifecycle and Monitoring12 topics
- Data, Code, Model, and Configuration Lineage
- Experiment Tracking and Model Registries
- Reproducible Training and Evaluation Pipelines
- Training-Serving Consistency
- Model Release Criteria and Review Evidence
- Shadow and Limited-Rollout Evaluation
- Data Drift and Prediction Drift
- Delayed Labels and Performance Monitoring
- Model Degradation and Retraining Triggers
- Feedback Loops After Deployment
- Rollback and Model Retirement
- Model Cards and Operational Documentation
89Uncertainty, Robustness, and Distribution Shift12 topics
- Aleatoric and Epistemic Uncertainty
- Deep Ensembles and Approximate Bayesian Methods
- Calibration Under Distribution Changes
- Conformal Prediction and Exchangeability Assumptions
- Prediction Sets and Coverage Evaluation
- Out-of-Distribution Detection
- Corruptions, Perturbations, and Natural Variation
- Spurious Correlations and Shortcut Dependence
- Distributionally Robust Objectives
- Abstention and Escalation Strategies
- Stress Testing Across Plausible Deployment Conditions
- Distinguishing Confidence From Correctness
90Interpretability and Model Understanding12 topics
- Global and Local Explanations
- Intrinsically Interpretable Model Choices
- Feature Importance and Permutation Tests
- Partial Dependence and Correlated Inputs
- Local Surrogates and Attribution Methods
- Gradient-Based Explanations
- Counterfactual Explanations and Feasibility
- Concept-Based Interpretation
- Probing Learned Representations
- Mechanistic Interpretability and Activation Interventions
- Faithfulness, Stability, and Explanation Evaluation
- Limits of Interpreting Model-Generated Explanations
91Fairness and Human Impact12 topics
- Stakeholders and Affected Populations
- Sources of Bias Across the Model Lifecycle
- Group and Individual Fairness Concepts
- Fairness Metrics and Incompatible Objectives
- Subgroup Performance and Intersectional Analysis
- Bias Mitigation Through Data and Training Choices
- Accessibility and Unequal Error Burdens
- Human Review and Decision Authority
- Automation Bias and Overreliance
- Contestability and Correcting Model Errors
- Documenting Tradeoffs and Unresolved Limitations
- Evaluating Whether an AI System Should Be Used
92AI Security and Misuse Risk12 topics
- Threat Models for Learned Systems
- Adversarial Input and Evasion Risks
- Training Data Poisoning and Backdoors
- Model Extraction and Information Leakage
- Prompt Injection in Retrieval and Tool Workflows
- Untrusted Models, Datasets, and Dependencies
- Capability Boundaries and Least-Privilege Integration
- Misuse Evaluation and Defensive Testing
- Content Safeguards and Their Limitations
- Monitoring Abuse and Unexpected System Behavior
- Incident Response for Compromised AI Components
- Evidence-Based Release and Access Decisions
93Foundation Model Evaluation13 topics
- Capability-Specific and Application-Specific Test Sets
- Benchmark Validity and Data Contamination
- Factual Accuracy and Unsupported Claims
- Instruction Following and Constraint Satisfaction
- Multilingual and Cross-Domain Performance
- Long-Context Retrieval and Information Use
- Human Evaluation and Annotation Agreement
- Model-Based Judges and Judge Bias
- Pairwise Preferences and Ranking Uncertainty
- Robustness to Prompt and Context Variations
- Safety, Refusal, and Over-Refusal Evaluation
- Cost and Latency Alongside Quality
- Evaluation Changes After Adaptation and Compression
94AI Research and Reproducible Evidence12 topics
- Reading Research Claims and Experimental Designs
- Separating Contributions From Implementation Choices
- Reproducing Baselines Before Extensions
- Ablations and Controlled Comparisons
- Reporting Data, Compute, and Search Budgets
- Random Seeds and Result Variability
- Negative Results and Failed Hypotheses
- Benchmark Overfitting and Selective Reporting
- Dataset and Model Documentation
- Releasing Reproducible Artifacts
- Identifying Unsupported Generalization Claims
- Turning Research Results Into Engineering Decisions
Stage 7
Practicum & Capstone
Your own system, end to end
4 modules · 43 topics
95Classical AI and Predictive Learning Practicum10 topics
- Formulating a Search or Planning Problem
- Comparing Heuristics Under a Fixed Budget
- Building a Constraint-Based Solution
- Preparing a Supervised Learning Dataset
- Establishing Linear and Tree-Based Baselines
- Comparing Kernel and Ensemble Models
- Designing Leakage-Resistant Evaluation
- Tuning Models Under Resource Constraints
- Analyzing Calibration and Failure Slices
- Presenting a Reproducible Model Selection Report
96Deep and Generative Learning Practicum10 topics
- Implementing and Verifying a Neural Training Loop
- Comparing Neural Architecture Choices
- Training a Self-Supervised Representation
- Adapting a Pretrained Model to a Defined Task
- Building a Small Generative Model
- Constructing a Minimal Transformer Language Model
- Evaluating Data Quality and Training Stability
- Testing Generation Quality and Diversity
- Measuring Compression and Inference Tradeoffs
- Producing a Documented Deep Learning Experiment
97Decision-Making and Agent Evaluation Practicum10 topics
- Defining a Sequential Decision Environment
- Comparing Bandit or Reinforcement Learning Baselines
- Evaluating Policies on Held-Out Scenarios
- Building a Retrieval-Grounded Task Workflow
- Integrating Tools With Explicit Action Boundaries
- Testing Memory and Multi-Step Task Execution
- Simulating Tool Failures and Conflicting Evidence
- Measuring Task Success, Cost, and Intervention Needs
- Building an Agent Regression Suite
- Reporting Reliability Limits and Deployment Conditions
98Integrated Artificial Intelligence Engineering Project13 topics
- Defining the User Need and AI System Boundary
- Establishing Data Access and Evaluation Requirements
- Selecting Symbolic, Statistical, Neural, or Hybrid Methods
- Building Baselines and a Reproducible Training Workflow
- Developing the Model and Task-Specific Adaptation
- Designing Retrieval, Tools, or Decision Logic Where Needed
- Evaluating Quality, Uncertainty, and Failure Modes
- Testing Fairness, Security, and Human Oversight Requirements
- Optimizing Resource Use and Inference Performance
- Preparing Model Interfaces and Release Evidence
- Defining Monitoring, Update, and Rollback Procedures
- Documenting Data, Models, Assumptions, and Limitations
- Presenting a Complete AI System and Evaluation Package
Fifteen minutes.Every day.
- 1
A lesson fits a lunch break
One idea at a time, in short slides. A whole lesson takes about fifteen minutes.
- 2
Practice with instant feedback
Questions sit inside the lesson. Answer one and see right away whether you got it.
- 3
A streak that brings you back
A lesson a day keeps the streak alive. Small, steady sessions carry you through.
- 1
A lesson fits a lunch break
One idea at a time, in short slides. A whole lesson takes about fifteen minutes.
- 2
Practice with instant feedback
Questions sit inside the lesson. Answer one and see right away whether you got it.
- 3
A streak that brings you back
A lesson a day keeps the streak alive. Small, steady sessions carry you through.
Built for people who build.
Engineers moving into ML
Analysts ready to build models
Product teams shipping AI features
Where this course leads.
The job this course is built around, and how people get into it.
Machine Learning Engineer
Builds, trains and ships the models behind a product, and keeps them working after launch.
All future careersOn the job
- Turn a product question into a dataset and a model
- Evaluate a model honestly before it reaches users
- Put a model into production and watch it drift
How people get in
Usually a degree in computer science, maths or a related field, or a strong software background plus a portfolio of shipped models.
Be first in line.
Early access for individuals, pilots for teams. Tell us who's learning.
enterprise@astratrainer.com