tldr - powered by Generative AI

• Rootless containers can solve the issue of containers running as root, but there is a limitation in building OCI containers rootless
• The controllers in Kubeflow pipelines run as cluster admin, which is a security risk
• Namespace sharing can also be a security risk as collaborators gain access to service accounts
• Solutions proposed include reducing the complexity of controllers and using reduced cluster roles
• The presentation highlights the progress made in Kubeflow security, such as authentication and machine-to-machine authentication

tldr - powered by Generative AI

• Tokenization is used to protect patient privacy by changing personally identifiable information to a token based on a hardware security key.
• Strict confinement features of micro-kubernetes distribution are used to ensure tamper-proof tokenization.
• Confidential computing is used to expand local Kubernetes clusters in a safe way by creating a VM on a public cloud and utilizing open enclave and open source projects to configure the confidential compute and underlying hardware features.
• The benefits of using public clouds for research use cases are discussed, including the ability to spike up capacity when training a bigger model.
• The presentation emphasizes the importance of using secure MLOps to comply with industry standards and protect patient privacy.

tldr - powered by Generative AI

• Mig technology allows for efficient use of GPU resources by splitting one unit of GPU into multiple instances
• GPUdirect RDMA enables efficient distributed processing for deep learning tasks
• POC results show that Mig technology is suitable for model development and inference tasks, while GPUdirect RDMA is suitable for larger scale tasks
• Points to consider for Kubernetes testing are discussed in the presentation

tldr - powered by Generative AI

• Kubernetes can be used for running HPC workloads, but tuning and optimization are crucial
• OpenFOAM was used as an example of an MPI job that can be run on Kubernetes
• Future developments include using Nvidia GPUs and exploring new schedulers

tldr - powered by Generative AI

• Kubeflow Metal is a terraform module that deploys Kubeflow on a Kubernetes cluster on bare metal servers
• It is a cheaper alternative to cloud infrastructure with a fixed cost
• It allows for quick bootstrapping of an ML environment or infrastructure for a team
• Deployment is elastic and easily scalable
• It can be used for plugging into a CI/CD process
• It is useful for cases where data cannot be moved to the cloud, such as financial or insurance data
• Kubeflow Metal is looking for people to help improve the project

tldr - powered by Generative AI

• Q4 Pipelines proposes a new component registry to address problems with authoring, publishing, and maintaining components
• The registry will have a unified YAML format, versioning and tagging capabilities, and direct integration with the Q4 Pipelines SDK
• Third-party registries can also implement the server-side of the API
• The Machine Learning Exchange is an example of a registry that is implementing the new protocol and offers various asset types
• Watson Studio Pipelines is in open beta and provides a canvas for running experiments and integrating notebooks

tldr - powered by Generative AI

• Using machine learning can improve results in studying subatomic particles, as demonstrated by the jet energy regression example
• Kubeflow can help run machine learning workloads
• Challenges in implementing the demo included finding the correct version of the Triton server image and customizing TensorBoard
• Possible improvements include profile replication across multiple clusters, making pipelines namespace, and adding limit range resources to profiles

tldr - powered by Generative AI

• Kubernetes allows for elastic scaling but has limitations when it comes to fitting large matrices in memory
• Apache Spark and Mahout can distribute matrices across an unbounded number of pods/nodes
• Kubeflow can be used to make the process easily reproducible
• The presentation provides an anecdote about using these tools to denoise DICOM images of lungs of COVID patients

tldr - powered by Generative AI

• Investing in observability and reliability preemptively before experiencing issues
• Using infrastructure as code, specifically Terraform and Argo CD, to manage multi-cluster deployments and ensure consistency
• Creating custom metrics, such as Kubeflow state metrics, to track specific product needs and enable effective SLOs and alerts