December 10, 2019

439 words 3 mins read

Deep learning with TensorFlow and Spark using GPUs and Docker containers

Deep learning with TensorFlow and Spark using GPUs and Docker containers

In the past, you needed a high-end proprietary stack for advanced machine learning, but today, you can use open source machine learning and deep learning algorithms available with distributed computing technologies like Apache Spark and GPUs. Nanda Vijaydev and Thomas Phelan demonstrate how to deploy a TensorFlow and Spark with NVIDIA CUDA stack on Docker containers in a multitenant environment.
Talk Title Deep learning with TensorFlow and Spark using GPUs and Docker containers
Speakers Nanda Vijaydev (BlueData), Thomas Phelan (HPE BlueData)
Conference Strata Data Conference
Conf Tag Making Data Work
Location London, United Kingdom
Date May 22-24, 2018
URL Talk Page
Slides Talk Slides
Video

Organizations these days understand the need to keep pace with newer technologies and methodologies when it comes to doing data science with machine learning and deep learning. Data volume and complexity is increasing by the day, so it’s imperative that companies understand their business better and stay on par with or ahead of the competition. Thanks to products such as Apache Spark, H2O, and TensorFlow, these organizations no longer have to lock in to a specific vendor technology or proprietary solutions. These rich deep learning applications are available in the open source community, with many different algorithms and options for various use cases. However, one of the biggest challenges is how to get all these open source tools up and running in an easy and consistent manner (keeping in mind that some of them have OS kernel and software components). For example, TensorFlow can leverage NVIDIA GPU resources, but installing TensorFlow for GPU requires users to setup NVIDIA CUDA libraries on the machine and install and configure the TensorFlow application to make use of the GPU computing ability. The combination of device drivers, libraries, and software versions can be daunting and may be a nonstarter for many users. Since GPUs are a premium resource, organizations that want to leverage this capability need to bring up clusters with these resources on-demand and then relinquish their use after computation is complete. Docker containers can be used to set up this instant cluster provisioning and deprovisioning and can help ensure reproducible builds and easier deployment. Nanda Vijaydev and Thomas Phelan demonstrate how to deploy a TensorFlow and Spark with NVIDIA CUDA stack on Docker containers in a multitenant environment. Using GPU-based services with Docker containers does require some careful consideration, so Thomas and Nanda share best practices specifically related to the pros and cons of using NVIDIA-Docker versus regular Docker containers, CUDA library usage in Docker containers, Docker run parameters to pass GPU devices to containers, storing results for transient clusters, and integration with Spark. Topics include:

container apache gpu spark algorithm tensorflow complexity data science cuda open source deep learning machine learning docker nvidia use case cluster
comments powered by Disqus
Distributed deep learning with containers on heterogeneous GPU clusters

Distributed deep learning with containers on heterogeneous GPU clusters

November 26, 2019

Deep learning model performance relies on underlying data. Dong Meng offers an overview of a converged data platform that serves as a data infrastructure, providing a distributed filesystem, key-value storage and streams, and Kubernetes as orchestration layer to manage containers to train and deploy deep learning models using GPU clusters.
Distributed training of deep learning models

Distributed training of deep learning models

December 10, 2019

Mathew Salvaris, Miguel Gonzalez-Fierro, and Ilia Karmanov offer a comparison of two platforms for running distributed deep learning training in the cloud, using a ResNet network trained on the ImageNet dataset as an example. You'll examine the performance of each as the number of nodes scales and learn some tips and tricks as well as some pitfalls to watch out for.
Data science in the cloud

Data science in the cloud

November 27, 2019

In this talk Alex will discuss lessons learned from AWS SageMaker, an integrated framework for handling all stages of analysis. AWS uses open source components such as Jupyter, Docker containers, Python and well established deep learning frameworks such as Apache MxNet and TensorFlow for an easy to learn workflow.
Deep learning for domain-specific entity extraction from unstructured text

Deep learning for domain-specific entity extraction from unstructured text

November 27, 2019

Mohamed AbdelHady and Zoran Dzunic demonstrate how to build a domain-specific entity extraction system from unstructured text using deep learning. In the model, domain-specific word embedding vectors are trained on a Spark cluster using millions of PubMed abstracts and then used as features to train a LSTM recurrent neural network for entity extraction.
Scaling the AI hierarchy of needs with TensorFlow, Spark, and Hops

Scaling the AI hierarchy of needs with TensorFlow, Spark, and Hops

December 5, 2019

Distributed deep learning can increase the productivity of AI practitioners and reduce time to market for training models. Hadoop can fulfill a crucial role as a unified feature store and resource management platform for distributed deep learning. Jim Dowling offers an introduction to writing distributed DL applications, covering TensorFlow and Apache Spark frameworks that make distribution easy.
Machine-learned model quality monitoring in fast data and streaming applications

Machine-learned model quality monitoring in fast data and streaming applications

December 7, 2019

Most machine learning algorithms are designed to work on stationary data, but real-life streaming data is rarely stationary. Models lose prediction accuracy over time if they are not retrained. Without model quality monitoring, retraining decisions are suboptimal and costly. Emre Velipasaoglu reviews monitoring methods, focusing on their applicability in fast data and streaming applications.