This week during the re:Invent conference, AWS announced a series of new releases that bring its SageMaker platform closer to the needs of real world machine learning solutions.Before we dig deeper into SageMaker’s new capabilities, let’s understand the statement about the gap between the capabilities of cloud AI platforms and the requirements of real world machine learning applications..Areas such as training, data labeling, integration with the latest AI research, portability, optimization end up consuming a lot of time and resources and remain very difficult hurdles for data science teams.Almost since its first release, SageMaker has become the platform of choice for machine learning technologist in AWS..As a result, data science teams spend considerable amounts of time collecting and labeling datasets so that they can be used to train machine learning models..These data labeling exercises are typically disconnected from other aspects of the machine learning lifecycle such as training or implementation.SageMaker Ground Truth brings elastically scalable data labeling capabilities of SageMaker applications..At a high level, SageMaker Ground Truth processes consists of four fundamental steps:Store your data in Amazon S3,Create a labeling workforce,Create a labeling job,Get to work,Visualize results.Ground Truth expands SageMaker into very early stages of machine learning workflows when data is being collected and labeled.Real Reinforcement Learning Using SageMaker RLReinforcement learning(RL) has been at the center of some of the most recent breakthroughs in AI such as AlphaGo or OpenAI Five..My team at Invector Labs is already hacking thru this.Train Once and Run Everywhere Using SageMaker NeoOptimizing models to run on specific hardware configuration is one of the most painful exercises of machine learning solutions..SageMaker Neo also provides a runtime for each target platform that loads and executes the compiled model.Intelligent Compute Acceleration using SageMaker Elastic InferenceScaling machine learning models is mostly a subjective exercise.. More details