AWS Machine Learning Specialty
Demonstrate expertise in building, training, tuning, and deploying ML models using the AWS cloud.
Exam Details
Your Progress
0%0 of 8 steps completed
Step-by-Step RoadmapEstimated total: 10 to 16 weeks
Prerequisites: AWS Cloud Practitioner + ML fundamentals
This is an AWS specialty cert. You need solid AWS familiarity AND ML knowledge before you start.
What to Learn
- AWS Cloud Practitioner level knowledge
- Python and basic data science
- Pandas, NumPy basics
- Linear regression, classification basics
- Cross-validation, train/test split
Resources
- AWS CLF-C02 study path on PathCert
- Andrew Ng's Machine Learning Coursera course
- Kaggle Learn (free)
If you cannot explain bias-variance trade-off and confusion matrix, you are not ready. Spend extra time on ML fundamentals before AWS specifics.
Data engineering on AWS
How data gets into AWS for ML: S3, Glue, Kinesis, ingestion patterns.
What to Learn
- S3 for ML data storage
- AWS Glue for ETL
- Kinesis for streaming data
- AWS Data Pipeline
- Data formats: CSV, Parquet, RecordIO
Resources
- Stephane Maarek AWS ML Udemy course
- AWS Whitepaper: ML Lens
Format choice matters: RecordIO/Protobuf for SageMaker built-in algorithms. Parquet for analytics. Know when to use each.
Exploratory data analysis
Cleaning, transforming, visualising data on AWS. Feature engineering.
What to Learn
- Missing data handling
- Feature scaling (standardisation, normalisation)
- Encoding categorical variables (one-hot, label, target)
- Outlier detection and treatment
- Feature engineering techniques
- EDA with Amazon SageMaker Data Wrangler
Resources
- Stephane Maarek course
- Hands-on with SageMaker free tier
Many exam questions are about which transformation to use. Memorise: standardisation for linear models and neural networks; tree-based models are scale-invariant.
Modelling: algorithms and SageMaker built-ins
Pick the right algorithm. Train it on SageMaker.
What to Learn
- SageMaker built-in algorithms: XGBoost, Linear Learner, K-Means, K-NN, BlazingText, Object2Vec, DeepAR, Random Cut Forest, IP Insights
- Algorithm choice by problem type
- Hyperparameter tuning
- SageMaker training jobs and instance types
Resources
- AWS docs on SageMaker built-in algorithms
- Stephane Maarek's algorithm cheat sheet
Memorise which algorithms solve which problems. The exam tests this constantly: which algorithm for time-series forecasting (DeepAR)? Which for anomaly detection (Random Cut Forest)? Which for text classification (BlazingText)?
Deep learning on AWS
Neural networks, TensorFlow/PyTorch on SageMaker, transfer learning.
What to Learn
- Neural network basics
- CNN for images, RNN/LSTM for sequences
- SageMaker script mode (custom code with TF/PyTorch)
- Transfer learning concepts
- GPU instance types (P3, P4, G4, G5)
Resources
- DeepLearning.AI specialisations (free to audit)
- SageMaker example notebooks
Know which instance types are GPU (P, G) vs CPU. Multi-GPU training options. Spot instances for cost savings on training.
Implementation, operations, deployment
Deploy models to production. Monitor them. Handle drift.
What to Learn
- SageMaker endpoints (real-time, batch transform, async)
- Multi-model endpoints
- A/B testing model variants
- Model monitor for drift
- Inference pipelines
- Edge deployment with SageMaker Neo
Resources
- AWS ML Specialty official sample questions
- SageMaker documentation
Know the trade-offs: real-time endpoints (low latency, costly), batch transform (cheap, no latency requirement), async (long-running, large payloads). Most candidates only know real-time.
AI/ML AWS services beyond SageMaker
Rekognition, Comprehend, Translate, Polly, Transcribe, Lex, Forecast, Personalize, Textract.
What to Learn
- Amazon Rekognition (computer vision)
- Comprehend (NLP)
- Translate, Polly (TTS), Transcribe (STT), Lex (chatbots)
- Forecast (time-series), Personalize (recommendations), Textract (OCR)
Resources
- AWS service one-pagers
Easy questions: 'which AWS AI service for X use case'. Memorise the lookup table: image = Rekognition, sentiment = Comprehend, recommendations = Personalize, etc.
Practice tests and final review
Take multiple full practice exams. Review every wrong answer.
What to Learn
- Tutorials Dojo AWS MLS practice tests (highly recommended)
- Stephane Maarek practice tests
- Official AWS sample questions
- Re-read all whitepapers
Resources
- Tutorials Dojo Jon Bonso AWS MLS tests
- Whizlabs AWS MLS
AWS MLS is one of the hardest specialty exams. Score 85%+ on Tutorials Dojo before booking. Most candidates underestimate this exam and have to retake.