Successfully reported this slideshow.
We use your LinkedIn profile and activity data to personalize ads and to show you more relevant ads. You can change your ad preferences anytime.

Machine Learning made easy


Published on

Cloud Native Night February 2019, Munich: Talk by Ian Schröder & Keith Tenzer (Red Hat)

Join our Meetup:

Abstract: Machine Learning does not have to be a pain point for public cloud providers or crazy coders. With a tight framework consisting of an enterprise open source solution and a smart container platform, it can be straightforward and even painless.
Including a demo of "Machine Learning made easy" you will soon see the power of a smart integrated platform. Red Hat delivers a demo showing how to build a framework for Machine Learning of a voice to text solution.

Published in: Data & Analytics
  • Be the first to comment

Machine Learning made easy

  1. 1. Cloud Native Night Feb/21 A coding demo "Machine Learning made easy” Ian Schröder, Solution Specialist Middleware, Keith Tenzer, Solutions Architect, 21 February 2019
  2. 2. AGENDA 2 What is Machine Learning? 3 important Points DEMO
  3. 3. 3 ABOUT KEITH AND IAN 20 Linux/Storage Admin, Developer, Architect 1 15 Joined Red Hat 47 Years experience working with people 1 18 Joined Red Hat
  4. 4. 4 WHAT IS MACHINE LEARNING? Deep Learning Machine Learning Artificial Intelligence Algorithms that parse data, learn from the data and then apply learning to make informed decisions. A broad field where machines are cognitive, able to analyze environment in order to make decisions. A subset of machine learning, algorithms can determine on their own if prediction is accurate or not, meaning they learn without supervision.
  5. 5. 5 ● Machine learning is open for everyone ○ Machine Learning is not a proprietary offering from the major cloud providers ● Machine learning needs access to Data and Runtimes Take control with Container ○ Agile microservice frameworks, Apps and runtimes offers flexibility ○ Cloud-to-Cloud and On-Premise, mash your own network ○ Connect data-lakes to find information in your data ● Machine learning can be easy THREE IMPORTANT POINTS
  7. 7. 7 CONTAINER MAKE ML easy Python Compilers Polyglot Applications Datasets Models CI/CD GPUs Lots of memory Lots of Data ML Frameworks Lots of decoupled moving parts API´s
  9. 9. 9 HOW MACHINE LEARNING WORKS Choose Use Case Categorize Data Label Data Setup layers (neural network) Accuracy, optimize and training metrics Compile model Train model with training dataset Evaluate accuracyImprove Training Cycle Application Data Ingestion Analyze dataset based on model Predictions, Findings and Decisions Run Cycle
  10. 10. DEMO “Build a Siri/Echo/Alexa”
  11. 11. 11 MACHINE LEARNING AREA 2 START Anomaly Detection = Predictive
  12. 12. 12 ● GitHub Repository - ● Blog - ne-learning/ RESOURCES
  13. 13. Q&A
  14. 14. Keith Tenzer Senior Solution Architect Red Hat GmbH +49 151 44051401 Ian Schroeder Solution Specialist Middleware Red Hat GmbH +49 172 8471401
  16. 16. 16 DEMO APPLICATION ROLLOUT PodContainer Fedora Base C++ Dev tools Python + devel Python modules Nodejs BUILD DEPLOY RUN Container Fedora Base C++ Dev tools Python + devel Python modules Nodejs /deepspeech Download model Start Nodejs 8080 Start Script Registry OpenShift
  17. 17. 17 OPENSHIFT: OUR FOUNDATION FOR AI / MLLEVERAGE OPTIMAL CLOUD-NATIVE APP DEV CAPABILITIES AWS Microsoft Azure Google CloudOpenStackDatacenterLaptop APPLICATION LIFECYCLE MANAGEMENT ENTERPRISE CONTAINER HOST (RHEL) CONTAINER ORCHESTRATION AND MANAGEMENT (KUBERNETES) Much more than just RHEL + Docker + K8S: ● Security + CI/CD pipelines + hybrid cloud management + container-native storage + networking ● Microservices infrastructure: Istio service mesh for routing & traffic control, security, availability, and identity services ● Certified plugin/interoperability with leading storage, network vendors ● Available & optimized for private & public clouds in self-managed or RHT managed offerings ● Fully integrated with RHT middleware platforms & services
  18. 18. 18 GPU SUPPORT IN OPENSHIFT ● Joint collaboration with strategic partners for drivers, plugins and container images ● Device Manager GA ● Scheduler: Priority and preemption ● Seamless install experience of drivers, plugins and dependencies ● Container images in RHCC/ISV Registry ● Certifications and support RHEL Host Device Manager plugin Kublet Device Manager Kubernetes SchedulerRHEL base image + Vendor libraries for GPUs + Frameworks for AI/ML such as Tensorflow or Pytorch Device drivers for GPU
  19. 19. 19 Upstream Community Project with Google, RedHat, Microsoft, Intel, Caicloud and others Democratizing AI with Machine Learning for Everyone Challenges in ML that Kubeflow is addressing: Composability, Portability, Scalability KUBEFLOW
  20. 20. 20 End-to-End Kubeflow Workflow Experiment with Jupyter Deploy Kubeflow on OpenShift Deploy TensorFlow, PyTorch, Caffe2, Horovod to build model Katib for Hyperparameter Tuning Kubeflow Volume Controller Model Serving and Inference with TensofFlowServing or SeldonCore Data Scientists S3 Object Store Build container image of model Integrate model with your application Operators App Developers Consumers Argo for workflow design