Home AI News Discover the Latest AI/ML Services at re:Invent 2018

Discover the Latest AI/ML Services at re:Invent 2018

Introduction
The ML Stack at AWS
- Infrastructure and Framework Layer
- Services and Platform Layer
- AI Applications Layer
Themes at Reinvent 2018
- Speed and Cost
- Data and Data Labeling
- Making ML Easier to Use
New Managed AI Services
- Amazon Forecast
- Personalize
- Amazon Textract
- Amazon Comprehend Medical
Scaling Machine Learning
- Amazon Elastic Inference
- AWS Neo Compiler
Marketplace for Machine Learning
SageMaker Reinforcement Learning
- Using Scoring Function for Training
- DeepRacer and Reinforcement Learning
SageMaker Ground Truth for Data Labeling
Conclusion

Introduction

In this article, we will explore the latest advancements and services in the field of machine learning (ML) provided by AWS. We will discuss the ML stack at AWS, the themes that emerged during the recent Reinvent 2018 event, and the new managed AI services introduced by AWS. We will also dive into the topic of scaling machine learning, the marketplace for ML algorithms, and the SageMaker reinforcement learning capabilities. Lastly, we will explore SageMaker Ground Truth for data labeling. So, let's get started!

The ML Stack at AWS

The ML stack at AWS can be divided into three layers: the infrastructure and framework layer, the services and platform layer, and the AI applications layer.

Infrastructure and Framework Layer

At the bottom layer of the ML stack, we have the infrastructure and framework layer. This layer focuses on providing the necessary infrastructure and frameworks for data scientists to work with. It includes accelerated hardware instances such as GPU and FPGA instances, as well as frameworks like TensorFlow, MXNet, and PyTorch.

Services and Platform Layer

Moving up a layer, we have the services and platform layer. This layer provides a range of services to assist data scientists, data engineers, and other stakeholders in their ML projects. This includes services like SageMaker, which is a powerful service for data scientists, as well as services for big data processing, such as EMR and Spark.

AI Applications Layer

At the top of the stack, we have the AI applications layer. The goal of this layer is to make ML capabilities accessible to any developer, regardless of their expertise in the field. AWS has developed a set of services across various domains, including vision, speech, language, and chatbots, to enable developers to incorporate machine learning capabilities into their applications without the need to build and train their own models.

Themes at Reinvent 2018

During the recent Reinvent 2018 event, several themes emerged in the field of machine learning. These themes included speed and cost, data and data labeling, and making ML easier to use. Let's explore each of these themes in more detail.

Speed and Cost

One of the key themes at Reinvent 2018 was the focus on improving the speed and cost efficiency of ML training and inference processes. While training models often receive the most attention, it is equally important to optimize the inference stage, as this is where a model's usefulness lies. AWS introduced several services and instance types, such as the P3d and Elastic Inference, to enhance the performance and cost-effectiveness of training and inference processes.

Data and Data Labeling

Another prominent theme at Reinvent 2018 was the importance of data and data labeling in machine learning projects. Labeled data is crucial for Supervised techniques, but obtaining accurate and sufficient labeled data can be a costly and time-consuming process. AWS introduced services like Amazon Forecast, Personalize, Textract, and Comprehend Medical to address the challenges associated with data labeling and processing in specific domains.

Making ML Easier to Use

The third theme that emerged at Reinvent 2018 was the focus on making machine learning more accessible and easier to use for developers. AWS introduced services like SageMaker, which simplifies the entire ML workflow, from data preparation to model training and inference. The Marketplace for Machine Learning also allows developers to access and use a wide range of pre-built ML algorithms to accelerate their projects.

New Managed AI Services

During Reinvent 2018, AWS introduced several new managed AI services aimed at simplifying ML workflows and providing developers with powerful capabilities. Let's explore these services in detail.

Amazon Forecast

Amazon Forecast is a managed service for making predictions based on time series data. It uses advanced algorithms and allows data scientists to leverage historical and explanatory data to generate accurate forecasts. With built-in algorithms and automatic model selection capabilities, Amazon Forecast eliminates the need for manual training, optimizing the forecasting process. It offers improved accuracy and cost savings, making it an ideal solution for businesses requiring accurate predictions.

Personalize

Personalize is a recommendation service that enables developers to provide personalized recommendations to their customers. It leverages machine learning algorithms to analyze user activity and preferences, generating personalized recommendations for various content types. With Personalize, developers can incorporate powerful recommendation capabilities into their applications without the need to build and train their own models. It offers flexibility and accuracy while reducing development time and costs.

Amazon Textract

Amazon Textract is an optical character recognition (OCR) service that goes beyond traditional OCR solutions. It not only extracts text from documents but also understands the structure and context of the extracted data. With the ability to identify data labels and provide confidence scores, Amazon Textract enhances the accuracy and efficiency of Data Extraction and processing. It eliminates the need for manual data extraction, making document processing faster and cost-effective.

Amazon Comprehend Medical

Amazon Comprehend Medical is a specialized service for processing medical data. It offers natural language processing (NLP) capabilities specifically designed for the medical domain. It can extract medical entities, relationships, and protected health information (PHI) from unstructured medical text, making it easier to analyze and understand medical data. With its high accuracy and focus on PHI protection, Amazon Comprehend Medical is a valuable tool for Healthcare providers and researchers.

Scaling Machine Learning

Scaling machine learning is a critical challenge for organizations as they Seek to process and analyze larger volumes of data. AWS provides solutions to address this challenge and optimize machine learning operations.

Amazon Elastic Inference

Amazon Elastic Inference is a service that allows users to attach GPU acceleration to Amazon EC2 instances, enabling cost-effective inference without the need for dedicated GPU instances. By decoupling inference from training, users can optimize costs and Scale inference independently, improving performance and reducing latency.

AWS Neo Compiler

AWS Neo Compiler is a compiler that optimizes machine learning models created in popular frameworks like TensorFlow and MXNet for specific hardware and environments. It enables models to run efficiently on constrained devices, such as those used in manufacturing and IoT applications. By minimizing the complexity and size of models, the Neo Compiler improves performance while maintaining accuracy.

Marketplace for Machine Learning

The Marketplace for Machine Learning in SageMaker allows developers to access a wide range of pre-built machine learning algorithms from trusted vendors. These algorithms cover various domains and use cases, providing developers with the flexibility to find and utilize the most appropriate algorithms for their projects. The marketplace simplifies the process of accessing and integrating algorithms, saving time and effort in developing ML solutions.

SageMaker Reinforcement Learning

SageMaker Reinforcement Learning is a powerful feature that allows developers to train models using reinforcement learning techniques. With the ability to learn from data and optimize behavior without relying on labeled data, reinforcement learning is ideal for complex environments with numerous states. SageMaker provides tools and frameworks for utilizing reinforcement learning, such as TensorFlow and MXNet, to train models in simulation environments and real-world applications.

SageMaker Ground Truth for Data Labeling

SageMaker Ground Truth is a feature designed to make data labeling more efficient and accurate. It combines human involvement and a semi-supervised approach to generate high-quality labeled data sets. By starting with a small amount of labeled data and utilizing automatic labeling within a confidence threshold, users can achieve efficient data labeling. However, human involvement is essential for quality control, ensuring accurate labeling.

Conclusion

In conclusion, AWS continues to innovate and provide state-of-the-art services in the field of machine learning. The ML stack at AWS encompasses infrastructure and framework layers, services and platform layers, and AI applications layers. Themes at Reinvent 2018 revolved around speed and cost, data and data labeling, and making ML easier to use. AWS introduced new managed AI services, such as Amazon Forecast, Personalize, Textract, and Comprehend Medical, to address the challenges associated with ML workflows. Additionally, AWS offers solutions for scaling machine learning, a marketplace for accessing pre-built ML algorithms, reinforcement learning capabilities in SageMaker, and data labeling efficiency with SageMaker Ground Truth. With these advancements and services, developers can leverage AWS to accelerate their ML projects and achieve accurate and efficient results.

Resources: