AWS Sagemaker

These are some notes about the basis of Sagemaker

Sagemaker services

SageMaker Neo optimizes the trained model and compiles it into an executable. Taking the target hardware where the model will be run as input; the compiler uses a ML model to apply performance optimizations on your model.

Ground truth makes easy to label data. It has templates for common labeling tasks.

Rekognition is an artificial intelligence service that enables the image and facial recognition. It’s a fully managed service based on deep learning. It offers an API to analyze images.

DeepLens is a pack composed by:

  • A video camera with storage and CPU that enables to record videos and run a ML model to detect objects, recognize activity or other things.
  • A software package that enables the ML learner to build its own solutions based on Sagemaker and Lamda.
  • Templates available for: object detection, activity recognition, face detection, head pose detection, cat / dog….

Polly translates text information into speech. This is an end to end text-to-speech service.

  • Right now supports 24 languages with different voices.
  • Uses Speech Synthesis Markup Language (SSML), which is a XML based mark up language.
  • You can use it for content creation, test in-game dialogs, generated speech as voice response.
  • With lexicons we can define a specific dictionary for specific topics.
  • You can add different rates to the text to speech with the mark up language.

Lex powers Amazon Alexa. This service enables conversational interfaces using voice and text. It includes automatic speech recognition (ASR) and Natural language understanding (NLU) to try to recognize the intent of the speech. Lex is not only available for Alexa, but for any developers that wants to build specific solutions.

Transcribe automatic speech-to-text recognition service. It does just the opposite to Polly. It can be continually trained. The result text contains a “confidence” level to enable the manual review for specific cases.

Translate enables the translation of text from one language to other.

  • This service encodes the input, take into account the semantic representation, attention mechanism and finally decodes the output.
  • For instance, Amazon uses it for the customer reviews of products done in different languages.
  • The service divides the text into paragraphs. It uses a neural network model.

Comprehend is a natural-language processing (NLP) service that uses machine learning to uncover valuable insights and connections in text. If you give a sentence, it can provide you some of these capabilities:

  • Sentiment detection (positive / negative)
  • Entities (organization, location, date, person…)
  • Language detection
  • Key phrases
  • Topic modeling

Some real examples: voice of customer analytics, semantic search, knowledge management / discovery.

Comprehend medical focused on the structure and extract valuable patient-care information. For a medical document, comprehend medical highlight key phrases and words that should be noticed by the reader.

Forecast is the service that predicts future points in a time series based on historical data.

  • Incorporate additional inputs that enable to improve the accuracy of the forecasted results.
  • Automates feature engineering, algorithm selection and model tuning.
  • Build on top of Sagemaker, it enables to incorporate algorithms from Sagemaker.

Elastic Inference (EI) Inference is run more frequently and the number of executions can be a high consumer in terms of cost. Elastic inference works on the optimization of the inference process, providing a cost saving point for the execution of the models in production.

  • EI lowers inference costs by up to 75%

Marketplace, Machine Learning section, this section of AWS marketplace offers ready to use algorithms used by other users (free and paid).

Leave a Comment