# Training and Serving ML workloads with Kubeflow at CERN

- Institution: CERN
- Department: Conferences
- Speaker: Dejan Golubovic
- Year: 2021
- Language: English
- Subjects: Physics, Computer science, Earth and environment
- Source: [CERN](https://videos.cern.ch/record/3008231)
- Licence: [CERN](http://copyright.web.cern.ch)
- Who can watch: Tweens profiles and older (sources give no ratings; placed by Invisible Network from the source's own words)

Machine Learning (ML) has been growing in popularity in multiple areas and groups at CERN, covering fast simulation, tracking, anomaly detection, among many others. We describe a new service available at CERN, based on Kubeflow and managing the full ML lifecycle: data preparation and interactive analysis, large scale distributed model training and model serving. We cover specific features available for hyper-parameter tuning and model metadata management, as well as infrastructure details to integrate accelerators and external resources. We also present results and a cost evaluation from scaling out a popular ML use case using public cloud resources, achieving close to linear scaling when using a large number of GPUs..

0:00:00 Slide 1

0:01:19 Slide 2

0:02:38 Slide 3

0:03:57 Slide 4

0:05:16 Slide 5

0:06:35 Slide 6

0:07:54 Slide 7

0:09:13 Slide 8

0:10:32 Slide 9

0:11:51 Slide 10

0:13:10 Slide 11

## Lectures

1. [Training and Serving ML workloads with Kubeflow at CERN](https://invisiblenetwork.ai/learn/cern/3008231/c6bae5f6014a), 14 min (CERN's player)