AWS Certified Machine Learning Specialty: Modeling Competency

  • 34m
  • 34 questions
The AWS Certified Machine Learning Specialty: Modeling Competency benchmark measures your ability to frame business problems as machine learning problems, select and train appropriate machine learning models for the problem, perform hyperparameter optimizations for tuning the model, and evaluate the models. A learner who scores high on this benchmark demonstrates that they have the skills necessary to build and develop machine learning solutions on AWS.

Topics covered

  • build and train an image classification model in SageMaker
  • build machine learning (ML) solutions by selecting existing resources and launch them with a single click in SageMaker Studio
  • create a SageMaker notebook to train and finetune an object detection algorithm
  • define the importance of the availability of good data and data pipeline design
  • define the key characteristics of good machine learning problems
  • describe how object detection algorithms built on top of VGG and ResNet work to predict the objects present in the image and their confident score
  • describe how to design a good output for a business problem
  • describe how to use SageMaker's€™s Object2Vec algorithm that learns low dimensional embeddings of high dimensional objects
  • describe hyperparameter tuning jobs in SageMaker and name recommended practices
  • describe SageMaker seq2seq algorithm that takes in a sequence and generates a sequence suitable for a range of tasks
  • describe the methodology behind principal component analysis (PCA) and the next level of linear learner
  • describe the most challenging problems in machine learning (ML)
  • evaluate the learning ability of a machine learning model and identify potential risks and biases in the dataset as well as their resulting impact
  • identify how to formulate a business problem into a machine learning problem
  • implement an anomaly detection system using Random Cut Forest in SageMaker
  • outline how supervised algorithms can be used to forecast time series based on past data
  • outline how to use Linear Learner and XGBoost (eXtreme Gradient Boosting) for classification and regression problems
  • outline how to use SageMaker's most simple classification/regression algorithm named K-NN and an unsupervised algorithm to find IP usage patterns
  • outline machine learning (ML) mindset and compare the ML approach to other problem-solving techniques
  • outline the basics of SageMaker's Neural Topic Model and Latent Dirichlet Allocation algorithms and list their primary use cases
  • practice reinforcement learning workflow with SageMaker
  • recognize how to complete clustering tasks in SageMaker
  • recognize the use of SageMaker's€™s semantic segmentation algorithm to predict the class of each pixel in an image and get shapes of objects
  • specify how to clearly define a business problem and set success and failure criteria
  • specify the factors that impact algorithm selection for a particular use case
  • use the integrated capabilities in SageMaker to connect EMR clusters with SageMaker Notebooks
  • work with Amazon CloudWatch to analyze real-time model performance by viewing training graphs of several performance metrics
  • work with SageMaker Clarify to analyze post-training bias of machine learning models
  • work with SageMaker Clarify to build explainable machine learning models
  • work with SageMaker Debugger to debug, monitor, and profile training jobs in real-time and reduce costs of your machine learning models by optimizing resources
  • work with SageMaker Experiments to organize, track, compare, and evaluate iterative machine learning experiments
  • work with SageMaker to implement PCA and K-means algorithm for image clustering
  • work with SageMaker to tune models over time and manage training and tuning costs by using Spot training
  • work with training Keras/Tensorflow models with SageMaker