At this week’s reInvent conference, Amazon Web Services (AWS), introduced several enhancements to SageMaker its machine-learning platform.
SageMaker Ground Truth will be available to all customers, the company announced. This service promises to reduce manual labor and costs associated labeling data for machine learning training.
SageMaker Ground Truth allows users to tap “human annotators” from third-party providers, the Amazon Mechanical Turk crowdsourcing service, or their own workforce to perform part of the labeling tasks. According to AWS’ announcement, the service uses the human-labeled data to guide the rest of the work. This can reduce the time required to complete the dataset and possibly cut costs by up to 70%.
Amazon SageMaker Ground Truth allows you to use active learning to automatically label your input data. In a blog post Julien Simon, AWS’ AI and machine-learning evangelist, explained that active learning is a machine-learning technique that identifies data which needs to be labeled manually and data that can easily be labeled automatically.
SageMaker RL, which is a reinforcement learning service, is also generally available. Reinforcement learning is a reward-based method of training a machine-learning model. Simon explained the concept in a blog post.
This is where a computer program (or an agent) interacts and interacts with its environment. Most of the time, it takes place in a simulator. An agent is awarded a reward for taking actions. The rewards are calculated by a user-defined function that outputs a numerical representation of the actions to be incentivized. The agent can learn the optimal strategy for decision-making by maximizing positive rewards.
SageMaker RL allows developers to integrate reinforcement learning into their processes. AWS announced that SageMaker RL “allows any developer build, train, deploy with reinforcement learning through managed reinforcement-learning algorithms, support for multiple frameworks (including Intel Coach, Ray RL), multiple simulation environment (including SimuLink, MatLab), and integration to AWS RoboMaker (AWS’s new robotics service which provides a simulation platform that integrates well into SageMaker RL.”
SageMaker Neo, a third SageMaker capability, is also available generally as of this week. SageMaker Neo is described by AWS as a “deep-learning model compiler [that] allows customers to train models once and then run them anywhere with up 2X performance improvement.”
SageMaker Neo supports many hardware platforms, including those from Arm and Intel, Nvidia, and top machine learning frameworks. AWS plans to open source SageMaker Ne “soon”, under the Apache license.
