Shashank Prasanna

As a deep learning developer, data scientist, or machine learning engineer, you can choose from multiple Amazon EC2 GPU instance types to meet the evolving requirements of training and deploying ML models at scale. You can access instances with different GPU memory sizes, NVIDIA GPU architectures, capabilities (precisions, TensorCores, NVLink), GPUs per instance, number of vCPUs, system...

From search and product recommendations to speech recognition and language translation, many services rely on machine learning (ML) to serve predictions. PyTorch is the deep learning framework of choice for many ML model developers.

AWS offers a breadth and depth of machine learning (ML) infrastructure you can use through either a do-it-yourself approach or a fully managed approach with Amazon SageMaker. In this session, explore how to choose the proper instance for ML inference based on latency and throughput requirements, model size and complexity, framework choice, and portability.

In this builders session, we show you how to set up your own machine learning environments and workflows on Amazon SageMaker together with AWS Deep Learning AMIs. First we explain how to use the prepackaged, optimized Amazon Machine Images (AMIs).

Deep learning software stacks can be complex to build, optimize, and maintain. With different versions of frameworks, libraries, runtimes, and drivers for CPUs and GPUs, developers and data scientists spend much time ensuring that the full software stack works well together during upgrades and system changes.

Data scientists and machine learning engineers use containers to create custom, lightweight environments to train and serve models at scale with deep learning frameworks, such as TensorFlow, Apache MXNet, and PyTorch, achieving consistency across development and deployment. In this chalk talk, we discuss how to use AWS Deep Learning Containers to train and serve models with Amazon SageMaker.

Time-series prediction—whether it is regression, classification, or anomaly detection—has experienced a renaissance in the past three years thanks to the application of neural models. In turn, AWS has released an open-source Gluon-based toolkit, GluonTS, for time-series prediction.