Incredible Support Vector Machine Edition

Agenda
*******
18:00 - 18:15; Reception
18:15 - 18:20; A welcome note from organizers & sponsors
18:20 - 18:50; Talk 1: Hans Ramsl (Weights & Biases) brings order to the Chaos with MLOps: You Are a Twin Or Live in the Andes? You Might be a Supermodel, Popular and Successful.
18:50 - 19:00; Q&A
19:00 - 19:30; Talk 2: Timea Magyar (Thinkport GmbH) talks about synthetic data generation with generative AI used in training plant leaf disease image classification models.
19:30 - 19:40; Q&A
19:40 - 19:50; Short break with refreshments
19:50 - 20:20; Talk 3: William Arias (GitLab) walks us through how GitLab trains and deploys ML models in the cloud.
20:20 - 20:35; Q&A
20:35 - 20:45; Wrap-up and Announcements
20:45 - 22:00; Socializing with food & drinks
22:00 End of the Event
******
Details:
TALK 1 - Hans Ramsl (Weights & Biases) brings order to the Chaos with MLOps: You Are a Twin Or Live in the Andes? You Might be a Supermodel, Popular and Successful.
Abstract:
In the swiftly changing realm of language models such as Gemini or Llama, akin to the fast-paced world of fashion supermodels, keeping track of emerging LLMs and foundational models poses a significant challenge. This talk explores the role of MLOps in bringing clarity, structure, and order to the chaos of model proliferation.
We discuss how MLOps practices can streamline the lifecycle management of language models, offering a systematic approach to deployment, monitoring, and maintenance. Additionally, we address the crucial question of defining and comparing "good" models, emphasizing the role of MLOps in systematic model evaluation.
Join us to discover how MLOps can empower practitioners to make informed decisions, stay ahead of the evolving landscape, and establish a sustainable and efficient model deployment ecosystem.

TALK 2 - Timea Magyar (Thinkport GmbH) talks about synthetic data generation with generative AI used in training plant leaf disease image classification models.
Abstract:
The talk includes a reference MLOps implementation with AWS Sagemaker and MLFlow and a small demo. Important components/aspects I would like to talk about in the above context are: the role of feature stores, experiments, model registries, model monitoring (including drift detection) and model explainability.

TALK 3 - William Arias (GitLab) walks us through how GitLab trains and deploys ML models in the cloud.
Abstract:
This presentation will cover the GitLab’s Data Science team’s approach to train, re-train, and deploy models to different environments leveraging cloud-native technologies and GitLab’s DevOps expertise. This presentation will demonstrate what has been reused from our DevOps expertise and what has been built to meet the new demands derived from the need to operationalize predictive and generative models.
The presentation will include a mix of GitLab’s architecture for automating model training and retraining using Kubernetes with GPU hardware to make it frictionless for data scientists working with these technologies and transparent for platform engineers.