Schedule: Platforms and infrastructure sessions

Join Denis Batalov for an overview of the Amazon SageMaker machine learning platform. Denis walks you through setting up an Amazon SageMaker notebook (a hosted Jupyter Notebook server), using a built-in SageMaker deep learning algorithm, and building your own neural network architecture using SageMaker's prebuilt TensorFlow containers. Read more.

Jonathan Hung, Keqiu Hu, and Anthony Hsu offer an overview of TensorFlow on YARN (TonY), a framework to natively run TensorFlow on Hadoop. TonY enables running TensorFlow distributed training as a new type of Hadoop application. Its native Hadoop connector, together with other features, aims to run TensorFlow jobs as reliably and flexibly as other first-class objects on Hadoop. Read more.

Nigel Toon explains how scaling IPUs will increase the productivity of machine intelligence researchers everywhere. Join in to explore what can we do and expect from the field with vastly more compute. Read more.

Diego Oppenheimer explains why machine learning is a natural fit for serverless computing, shares a general architecture for scalable ML, discusses issues he ran into when implementing on-demand scaling over GPU clusters at Algorithmia, and provides general solutions and a vision for the future of cloud-based ML. Read more.

Gaurav Kaul, Grigori Fursin, and Suneel Marthi share trade-offs and design choices that are applicable to deep learning models when training in the cloud, specifically focusing on convergence and numerical stability, which are very important for autonomous driving and medical imaging. They then demonstrate how to optimize cost, performance, and convergence using CPU spot instances in AWS. Read more.

Alan Nichol walks you through building fully machine learning-based voice and chatbots with the open source Rasa stack. Read more.

Mikio Braun looks back on the past 20 years of machine learning research to explore aspects of artificial intelligence. He then turns to current examples like autonomous cars and chatbots, putting together a mental model for a reference architecture for artificial intelligence systems. Read more.

The common perception of applying deep learning is that you take an open source or research model, train it on raw data, and deploy the result as a fully self-contained artifact. The reality is far more complex. Nick Pentreath shares lessons learned building a deep learning model exchange and discusses the future of standardized cross-framework deep learning model training and deployment. Read more.

Recent research has shown that training for quantization can lead to large gains in energy efficiency, and embedded runtime packages like TensorFlow Lite and Caffe2Go offer portability over a number of platforms. Cormac Brick asks, Why can't we have both performance and portability? Cormac explores industry challenges and details the progress needed to close the portability-performance gap. Read more.

SURFSara and Intel collaborated as part of the Intel Parallel Computing Center initiative to advance the state of large-scale neural network training on Intel Xeon CPU-based servers. Ananth Sankar, Valeriu Codreanu, Damian Podareanu, and Steve Smith share insights on several best-known methods for neural network training and present results from tests performed on Stanford's CheXNet project. Read more.

In recent years, we’ve seen a shift from traditional vision algorithms to deep neural network algorithms. While many companies expect to move to deep learning for some or all of their algorithms, they may have a significant investment in classical vision. Paul Brasnett explains how to express and adapt a classical vision algorithm to become a trainable DNN. Read more.

Christopher Cho details how to leverage Kubernetes and the mighty Kubernetes APIs to build a complete deep learning pipeline, from data ingestion and aggregation to preprocessing and ML training to serving. Along the way, Christopher covers Kubeflow, a Google open source solution for managing machine learning with TensorFlow in a portable, scalable manner. Read more.

Zhipeng Huang explains how resource representation (RR) works with various intermediate representation (IR) technologies to help achieve the democratization of AI. Read more.