Success: Thank you for
subscribing! You will begin receiving emails as new content is posted. You may unsubscribe any time by
clicking the link in the email.
Kubeflow news
Your weekly update of curated kubeflow news from across the web.
Still figuring out what is Kubeflow?
22 February 2021
Kubeflow has become quite popular in the MLOps community as the tool that enables data science teams to automate their workflows from data preprocessing to model deployment on Kubernetes.
However, with it’s made of many pieces, and while it keeps evolving, how can you effectively start using?
Learn Kubeflow from online courses
Started by Google, Kubeflow is a project which’s basics are presented on Coursera through a free training. During it, you will learn about
TensorFlow Extended (or TFX), which is Google’s production machine learning platform
How to automate your pipeline through continuous integration and continuous deployment
How to manage ML metadata
How to automate and reuse ML pipelines across multiple ML frameworks
Kubeflow training for the whole team
A possible fast-path, if you want to train all your team at once is Canonical’s offer of 4-day enterprise training. The training covers the following topics:
Machine Learning & Deep Learning Architecture
Introduction to Charmed Kubeflow, Canonical’s packaging of Kubeflow
Kubeflow Pipelines and components
MLOps and Advanced Topics
Labs
Note: Canonical’s full offer of services can be found here
ML models in production
Building models is a totally different story than putting them in production. This is why we found this guide into how Tensorflow Extended (TFX) can help you move your models effectively, going through the whole process. The tutorial is not only a dry presentation of the steps that you need to follow, but a proper use case that you can have into production by the end of it.
In June 2020, AWS introduced SageMaker components for Kubeflow. 6 months later, Antje Barth, Sr. developer advocate @AWS, presents how to build end-to-end ML workflows with Kubeflow Pipelines and how to leverage the benefits of Kubeflow Pipelines and SageMaker altogether.
AWS re:invents end-to-end ML workflows
Watch the video below:
If you are more curious, there is an entire stack of articles around SageMaker& Kubeflow
Lifen, the french platform for healthcare products, recently switched from Gitlab’s jobs to Kubeflow Pipelines for continuous learning capabilities and showcases the transition and its benefits.
Maciej Mazur, Product Manager @Canonical for Telco and AI/ML shares his insights on how to approach data science in the Telco industry, including how Kubeflow can be a key asset for innovation.
David Aronchick, co-founder of Kubeflow and head of OSS ML Strategy at Microsoft explains how you can use Kubeflow and MLOps to secure your AI/ML workloads.
Watch the video below:
Canonical Kubeflow Operators
5 November 2020
Canonical, the publisher of Ubuntu, announces Kubeflow operators and packages.
Within the last week, Canonical announced two new technologies that aim at improving the Kubeflow experience:
Charmed Kubeflow – A set of Kubeflow charm operators, that leverage Juju OLM technology for lifecycle management of the applications inside Kubeflow. Read the announcement.
Lightweight Kubeflow bundles – two new packages of pre-selected applications from the Kubeflow bundle to fit desktop (Kubeflow lite) and edge scenarios (Kubeflow edge). Read blog post.
Kubeflow 1.1 is out!
24 August 2020
New release, increased capabilities.
After 6 months since the release of 1.0, Kubeflow releases a new version with increased capabilities.
This new version has focused on improving ML Workflow Productivity, Isolation and Security, and GitOps. Here is a list of the enhanced features:
Fairing and Kale (Kubeflow Automated pipeLines Engine) for end-to-end workflows.
Katib hyperparameter tuning:
New frameworks & algorithms (goptuna with CMA-ES, DARTS, chocolate, hyperopt, skopt).
Flexible config & tuning options (new python SDK, experiments with undefined goal, new UI, new resume policy).
Install and operations to support GitOps, using blueprints and kpt primitives.
Isolation and security: multi-user Kubeflow pipelines, CVE scanning, and support for Google’s Private GKE and Anthos.
MXNet and XGBoost distributed training operators, simplify training on multiple nodes, and speeds model creation.
The Kubeflow 101 series of short videos is a great way to quickly get up to speed on Kubeflow concepts. This week, Stephanie Wong guides us through Hyperparameters and how you can use Katib to achieve the
Kubeflow with Amazon Sagemaker
Shashank Prasanna, Senior Developer Advocate at AWS walks viewers on how to get the best out of Kubeflow and Amazon Sagemaker, in the same stack.
Security breach
12 June 2020
Microsoft exposes attacks to Kubeflow deployments
Microsoft publishes report detailing series of attacks against clusters running Kubeflow with the purpose of mining cryptocurrencies. To ensure that you are on the safe side, follow the steps below:
1. When deploying Kubeflow, make sure that its dashboard isn’t exposed to the internet: check the type of the Istio ingress service by the following command and make sure that it is not a load balancer with a public IP:
kubectl get service istio-ingressgateway -n istio-system
2. Verify that the malicious container is not deployed in the cluster, through the following command:
kubectl get pods –all-namespaces -o jsonpath=”{.items[*].spec.containers[*].image}” | grep -i ddsfdfsaadfs
Amazon announced this week the possibility to configure Kubeflow Pipelines to run ML jobs with Amazon SageMaker.
This is yet another validation of Kubeflow as a widespread solution and reinforces the idea of ML workflows on Kubernetes. Read the post here.
Kubeflow 101
27 May 2020
Kubeflow in 3 minutes
Google launches Kubeflow 101, a series of short videos to demystify Kubeflow. Intro to Kubeflow. Watch an overview of Kubeflow in the first episode:
Intro to Kubeflow pipelines
In the second episode of Kubeflow 101, watch an introduction to Kubeflow pipelines, and the presentation of a new hosted version of Kubeflow
Weekend watch list
6 March 2020
CNCF introduction to Kubeflow
Arun Gupta, Senior Engineering Manager at Apple walks viewers through the benefits of Kubeflow and how to get started.
Securing Kubeflow
Talk by Barton Rhodes, Senior Machine Learning Engineer at DaVita Inc. This talk is motivated by Ian Coldwater during the Kubernetes Podcast ‘Attacking and Defending Kubernetes:
“Containers are only as secure as their runtimes, and their orchestration frameworks, and their kernels, and their operating systems, and everything else”
Building and monitoring Kubeflow
Kirk Kaiser, APM at Datadog, talks about how to build and monitor Kubeflow machine learning Services with examples for an art project.
