Trevor Grant

Working with big data matrices is challenging, Kubernetes allows users to elastically scale, but can only have a pod as large as a node, which may not be large enough to fit the matrix in memory. While Kubernetes allows for other paradigms on top of it which allows pods to coordinate on individual jobs, setting them up and making them play nice with ML platforms is not straightforward. Using Apache Spark and Apache Mahout we can work with matrices of any dimension and distribute them across an unbounded number of pods/nodes, and we can use Kubeflow to make our work quickly and easily reproducible. In this talk, we’ll discuss how we used Apache Spark and Mahout to denoise DICOM images of lungs of COVID patients and published our Pipeline with Kubeflow to make the process easily repeatable which could help doctors in more resource limited hospitals, as well as other researchers seeking to automate the detection of COVID.

Data Science, Machine Learning, and Artificial Intelligence has exploded in popularity in the last five years, but the nagging question remains, “How to put models into production?” Engineers are typically tasked to build one-off systems to serve predictions which must be maintained amid a quickly evolving back-end serving space which has evolved from single-machine, to custom clusters, to “serverless”, to Docker, to Kubernetes. In this talk, we present KubeFlow- an open source project which makes it easy for users to move models from laptop to ML Rig to training cluster to deployment. In this talk we will discuss, “What is KubeFlow?”, “why scalability is so critical for training and model deployment?”, and other topics.

Users can deploy models written in Python’s skearn, R, Tensorflow, Spark, and many more. The magic of Kubernetes allows data scientists to write models on their laptop, deploy to an ML-Rig, and then devOps can move that model into production with all of the bells and whistles such as monitoring, A/B tests, multi-arm bandits, and security.

Open source big data engines as well as HPC libraries seem to be proliferating at an increasing rate. Technical debt can be incurred with statistical and machine learning algorithms that require a highly specialized knowledge of the algorithm at hand as well as the distributed engine / HPC library which the method has been written against. The Apache Mahout project presents a highly modular stack which introduces levels of abstraction between the mathematical implementation of the algorithm (an R-Like Scala DSL) and the execution of the code. Users are able to interchange Apache Spark, Apache Flink (batch), and H2O distributed engines, as well as ViennaCL for OpenCL on GPU and OpenMP, and CUDA native solvers. Users can also port high level algorithms to new distributed engines or native solvers by defining a handful of BLAS operations.

Audience members will ideally have some concept of distributed engines such as Apache Spark, and a basic under standing of BLAS packs and linear algebra. (Basic understanding of linear algebra meaning they remember that things like matrix times matrix, matrix times vector, matrix transposed, and matrix decompositions are things that exist).

Current research is being done on creating a quantum BLAS pack for Apache Mahout, which will be a prototype of the next generation of High Performance Computing- however trying to even begin to explore the topic of quantum computing in 20 minutes is observed, and so research and progress at the time of the conference will be mentioned in passing only.