Learning path
Machine Learning learning path: everything to study, topic by topic
Comprehensive machine learning assessment covering statistics, classical ML, deep learning, MLOps, advanced techniques, and production ML systems.
Short answer
To learn Machine Learning, work through 6 modules and 29 topics in the order they build on each other. It starts with Foundations, then moves on to Classical ML and Deep Learning. Use the map as a checklist, or measure your level and study only what is missing.
What you will study
Topics are grouped by module. Open one to see its objectives and key concepts.
1 · Foundations4 topics
Statistics & Probability
Distributions, hypothesis testing, Bayes theorem, confidence intervals, and statistical significance
Objectives
- Apply Bayes theorem to classification problems
- Interpret hypothesis tests and confidence intervals
Key concepts
- probability distributions
- Bayes theorem
- hypothesis testing
- confidence intervals
- p-values
- central limit theorem
Linear Algebra
Vectors, matrices, eigendecomposition, SVD, and their role in ML algorithms
Objectives
- Explain how SVD applies to dimensionality reduction
- Compute and interpret matrix operations for ML
Key concepts
- matrix multiplication
- eigenvalues
- eigenvectors
- SVD
- dot product
- vector spaces
Data Preprocessing
Missing values, encoding, scaling, normalization, and data splitting strategies
Objectives
- Choose appropriate encoding for categorical variables
- Apply correct scaling strategies for different algorithms
Key concepts
- missing value imputation
- one-hot encoding
- label encoding
- min-max scaling
- standardization
- train-test split
Exploratory Data Analysis
Distribution analysis, correlation, outlier detection, feature relationships, and visualization
Objectives
- Identify data quality issues through EDA
- Select relevant features using correlation analysis
Key concepts
- correlation analysis
- outlier detection
- distribution plots
- feature distributions
- data visualization
- summary statistics
2 · Classical ML4 topics
Supervised Learning
Linear/logistic regression, decision trees, SVMs, k-NN, and bias-variance trade-off
Objectives
- Select appropriate supervised algorithms for given problems
- Explain the bias-variance trade-off and regularization
Key concepts
- linear regression
- logistic regression
- decision trees
- SVM
- k-NN
- bias-variance trade-off
- overfitting
Unsupervised Learning
K-means, hierarchical clustering, DBSCAN, PCA, and dimensionality reduction
Objectives
- Choose clustering algorithms based on data characteristics
- Apply PCA for dimensionality reduction effectively
Key concepts
- k-means
- hierarchical clustering
- DBSCAN
- PCA
- t-SNE
- silhouette score
- elbow method
Ensemble Methods
Bagging, boosting, random forests, gradient boosting, XGBoost, and stacking
Objectives
- Compare bagging and boosting strategies
- Tune gradient boosting hyperparameters effectively
Key concepts
- bagging
- boosting
- random forest
- gradient boosting
- XGBoost
- LightGBM
- stacking
Model Evaluation
Cross-validation, precision/recall, ROC-AUC, confusion matrix, and statistical tests
Objectives
- Select evaluation metrics aligned with business objectives
- Implement cross-validation to avoid data leakage
Key concepts
- cross-validation
- precision
- recall
- F1 score
- ROC-AUC
- confusion matrix
- stratified splits
3 · Deep Learning4 topics
Neural Networks
Perceptrons, backpropagation, activation functions, optimizers, and regularization
Objectives
- Explain backpropagation and gradient flow
- Choose appropriate activations and optimizers
Key concepts
- backpropagation
- activation functions
- SGD
- Adam optimizer
- dropout
- batch normalization
- vanishing gradients
Convolutional Neural Networks
Convolutions, pooling, architectures (ResNet, VGG), and computer vision applications
Objectives
- Design CNN architectures for image tasks
- Explain skip connections and their purpose
Key concepts
- convolution layers
- pooling
- stride and padding
- ResNet
- VGG
- feature maps
- receptive field
RNNs & LSTMs
Recurrent networks, LSTM gates, GRUs, sequence modeling, and attention basics
Objectives
- Explain LSTM gating mechanisms
- Choose between RNN variants for sequence tasks
Key concepts
- RNN
- LSTM gates
- GRU
- sequence-to-sequence
- vanishing gradients in RNNs
- bidirectional RNN
Transformers
Self-attention, multi-head attention, positional encoding, BERT, GPT, and fine-tuning
Objectives
- Explain the self-attention mechanism mathematically
- Compare encoder-only vs decoder-only architectures
Key concepts
- self-attention
- multi-head attention
- positional encoding
- BERT
- GPT
- encoder-decoder
- tokenization
4 · MLOps4 topics
Feature Engineering
Feature stores, feature extraction, selection, transformation pipelines, and data leakage
Objectives
- Design feature pipelines that prevent data leakage
- Implement feature stores for online/offline serving
Key concepts
- feature stores
- feature extraction
- feature selection
- data leakage
- feature pipelines
- online vs offline features
Model Serving
REST/gRPC endpoints, batch vs real-time inference, model versioning, and A/B testing
Objectives
- Design model serving infrastructure for production
- Implement safe model rollout with A/B testing
Key concepts
- REST inference API
- gRPC serving
- batch inference
- real-time inference
- model versioning
- A/B testing
- canary deployments
Experiment Tracking
MLflow, Weights & Biases, model registry, reproducibility, and metadata management
Objectives
- Set up experiment tracking for reproducible ML
- Manage model versions through a registry
Key concepts
- MLflow
- Weights & Biases
- model registry
- experiment reproducibility
- hyperparameter logging
- artifact tracking
Data Pipelines
ETL for ML, Airflow, data validation, drift detection, and pipeline orchestration
Objectives
- Design ML data pipelines with validation gates
- Implement drift detection for production models
Key concepts
- Airflow
- data validation
- data drift detection
- pipeline orchestration
- Great Expectations
- schema validation
5 · Advanced Techniques4 topics
Transfer Learning
Pre-trained models, fine-tuning strategies, domain adaptation, and few-shot learning
Objectives
- Design fine-tuning strategies for domain-specific tasks
- Apply parameter-efficient fine-tuning methods
Key concepts
- pre-trained models
- fine-tuning
- feature extraction
- domain adaptation
- few-shot learning
- LoRA
Reinforcement Learning
MDPs, Q-learning, policy gradient, actor-critic, and reward shaping
Objectives
- Formulate problems as MDPs with appropriate rewards
- Compare value-based and policy-based methods
Key concepts
- Markov decision process
- Q-learning
- policy gradient
- actor-critic
- reward shaping
- exploration vs exploitation
Generative Models
GANs, VAEs, diffusion models, autoregressive models, and evaluation metrics
Objectives
- Compare generative model architectures and trade-offs
- Diagnose training issues like mode collapse
Key concepts
- GANs
- VAEs
- diffusion models
- autoregressive generation
- FID score
- mode collapse
Hyperparameter Optimization
Grid search, Bayesian optimization, Optuna, early stopping, and search space design
Objectives
- Design efficient hyperparameter search strategies
- Apply Bayesian optimization for large search spaces
Key concepts
- grid search
- random search
- Bayesian optimization
- Optuna
- early stopping
- learning rate schedules
6 · Expert ML9 topics
Distributed Training
Data parallelism, model parallelism, gradient accumulation, and multi-GPU strategies
Objectives
- Design distributed training for billion-parameter models
- Choose parallelism strategies based on model size
Key concepts
- data parallelism
- model parallelism
- pipeline parallelism
- gradient accumulation
- all-reduce
- DeepSpeed
- FSDP
Interpretability
SHAP, LIME, attention visualization, feature importance, and fairness auditing
Objectives
- Apply SHAP/LIME for model explanations
- Audit models for bias and fairness issues
Key concepts
- SHAP values
- LIME
- attention visualization
- feature importance
- fairness metrics
- bias detection
ML System Design
End-to-end ML systems, latency budgets, feedback loops, and monitoring strategies
Objectives
- Design end-to-end ML systems with production constraints
- Build feedback loops for continuous model improvement
Key concepts
- end-to-end ML systems
- latency budgets
- feedback loops
- model monitoring
- data flywheel
- system trade-offs
Research to Production
Paper implementation, prototype to production, technical debt, and ML team processes
Objectives
- Bridge the gap between research prototypes and production
- Manage ML technical debt in large systems
Key concepts
- paper reproduction
- prototype to production
- ML technical debt
- testing ML systems
- team workflows
- experiment to deployment
Causal ML & Counterfactual Reasoning
Applying causal ML methods: double ML, uplift modeling, counterfactual estimation, and causal graphs in ML pipelines
Objectives
- Apply causal ML methods to estimate treatment effects in observational data
- Design uplift models for personalized intervention strategies
Key concepts
- double ML
- uplift modeling
- counterfactual estimation
- CATE
- causal graphs
- structural equations
- DoWhy framework
ML Robustness & Adversarial Examples
Understanding and defending against adversarial attacks, distribution shift, and model robustness evaluation
Objectives
- Evaluate model robustness under adversarial and distribution shift conditions
- Apply adversarial training to improve model robustness
Key concepts
- adversarial examples
- FGSM
- PGD
- adversarial training
- distribution shift
- robustness metrics
- certified defenses
Federated Learning Architecture
Designing federated learning systems: aggregation algorithms, privacy guarantees, communication efficiency, and production deployment
Objectives
- Design a federated learning system with differential privacy guarantees
- Address data heterogeneity and communication efficiency challenges in FL
Key concepts
- FedAvg
- differential privacy
- secure aggregation
- communication rounds
- heterogeneous data
- model poisoning
- TFF/PySyft
ML Interpretability at Depth
Applying advanced interpretability methods: SHAP values, LIME, integrated gradients, and model-agnostic explanations at scale
Objectives
- Select the appropriate interpretability method for a given model and stakeholder
- Scale SHAP explanations to production inference workloads
Key concepts
- SHAP
- LIME
- integrated gradients
- attention visualization
- counterfactual explanations
- global vs local explanations
- interpretability vs accuracy trade-off
Bayesian Deep Learning
Incorporating uncertainty quantification into deep learning: Bayesian neural nets, Monte Carlo dropout, and calibration
Objectives
- Implement uncertainty quantification for high-stakes ML predictions
- Calibrate deep learning models and evaluate calibration quality
Key concepts
- Bayesian neural networks
- MC dropout
- deep ensembles
- calibration (ECE)
- epistemic vs aleatoric uncertainty
- temperature scaling
- Laplace approximation
Levels: Junior, Mid, Senior, and Expert
Explain it in your own words
Graded by AI · no account · 1 questionNo options to guess from: you write, the AI grades against the topic's rubric and tells you what level your answer reads at.
Already studying Machine Learning?
The assessment measures topic by topic, and your plan skips what you already know.
Open to read. An account is only needed to be measured.
Frequently asked questions
Where should I start studying Machine Learning?
Start with Foundations: Statistics & Probability, Linear Algebra, and Data Preprocessing. The map orders topics by what they depend on, so each one prepares the next.
How long does a full Machine Learning assessment take?
A full assessment of this path takes about 60 minutes. Study time depends on where you start; a diagnostic shows which topics you can skip.
Do I need an account to use this page?
No. The map, objectives and key concepts are open. You only need an account to be measured, get your gap by topic and receive a plan.