Accueil › Certifications › MLS-C01 › Questions gratuites

Questions gratuites MLS-C01 — AWS Certified Machine Learning - Specialty

Téléchargez gratuitement 60 questions d'entraînement pour la certification MLS-C01 proposée par AWS. Toutes les questions sont accompagnées de corrections détaillées avec explications techniques.

Caractéristiques de l'examen blanc

Code de certificationMLS-C01
ÉditeurAWS
Nombre de questions60
TypeQCM avec 4 réponses possibles
Niveauassociate
CatégorieIT
Prix100% gratuit

Aperçu de 8 questions représentatives

Voici un échantillon aléatoire de 8 questions tirées de notre base d'entraînement MLS-C01. Pour accéder aux 60 questions complètes, lancez l'examen blanc gratuitement.

Question 1
Your training dataset requires feature scaling to ensure all features contribute equally. Which normalization techniques are appropriate?
  1. Min-Max scaling, standardization (z-score), or robust scaling
  2. Leave features at different scales
  3. Arbitrary scaling factors
  4. No feature scaling
Question 2
Your model training requires distributed training across multiple GPUs to reduce training time. Which SageMaker feature enables distributed training?
  1. SageMaker distributed training with data parallelism or model parallelism
  2. Single GPU training only
  3. Manual cluster setup
  4. Sequential single-node training
Question 3
Your ML pipeline requires consistent feature computation for training and inference. Which AWS service provides feature store capability?
  1. SageMaker Feature Store with online and offline stores
  2. Recompute features every time
  3. Store features in random databases
  4. No feature management
  5. , offline store (historical features in S3 for training via Athena), point-in-time correctness preventing data leakage (gets features as they existed at training time), and automatic feature metadata tracking. Feature Store ensures consistency between training and inference by using same feature computation code. Recomputing features wastes computation and risks inconsistency, random databases don't provide ML-specific capabilities (point-in-time correctness, feature metadata), and no feature management creates technical debt. Feature Store architecture: feature engineering pipeline (Glue/SageMaker Processing) → writes to Feature Store (online + offline) → training reads from offline store → inference reads from online store = consistent features with < 10ms latency. Essential for production ML at scale preventing train-serve skew and enabling feature reuse across teams.
Question 4
Your neural network training shows oscillating loss and fails to converge. Which hyperparameter adjustment helps convergence?
  1. Reduce learning rate, use learning rate scheduler, or change optimizer
  2. Increase learning rate significantly
  3. Remove all regularization
  4. Train for fewer epochs
Question 5
Your training data has privacy requirements restricting how it can be accessed. Which techniques enable privacy-preserving ML? (Choose TWO)
  1. Differential privacy adding noise to preserve privacy
  2. Federated learning training without centralizing data
  3. Share all raw data publicly
  4. No privacy protections
  5. Ignore privacy regulations
Question 6
Your ML project requires reproducibility of training runs. Which practices ensure reproducible training?
  1. Set random seeds, version data/code, track experiments, use SageMaker Pipelines
  2. Random initialization every time
  3. Undocumented code changes
  4. No experiment tracking
  5. , versioned code (Git SHA), logged hyperparameters, recorded training environment (Docker image), experiment tracking (SageMaker Experiments), automated pipelines (SageMaker Pipelines) = scientifically valid ML engineering. Critical for debugging, compliance, and knowledge sharing.
Question 7
Your model requires temporal features from time-series data. Which time-based features should be engineered? (Choose THREE)
  1. **Type** : multiple_choice
  2. Day of week, month, quarter
  3. Hour of day, is_weekend, is_holiday
  4. Time since last event
  5. Ignore temporal patterns
  6. Use timestamps as-is
Question 8
You need to train a deep learning model requiring GPU acceleration with managed infrastructure. Which SageMaker feature provides GPU training?
  1. SageMaker Training Jobs with GPU instance types (ml.p3, ml.p4)
  2. CPU-only training
  3. Manual EC2 GPU cluster management
  4. Desktop GPU training

Accédez aux 60 questions complètes gratuitement

Aucune carte bancaire requise. Examen chronométré, corrections détaillées, score final.

Lancer l'examen blanc MLS-C01 →

Pourquoi s'entraîner avec Certifexpress ?