Machine learning at scale

Macedonian Data Scientists, its a great pleasure to announce our 10th Data Science meetup. It will be held on the 19th of February 2020.

This meetup will feature 2 interesting talks on "Machine learning at scale" topic.

Snacks and drinks provided by Loka.
https://www.loka.com

Agenda:

• 17:45 - 18:15 - Gathering and networking (snacks and drinks)
• 18:15 - 18:50 - Anomaly Detection needs Attention (Sasho Nedelkoski)
• 18:50 - 19:25 - The good, the bad and the ugly in AutoML (Stefan Kochev)
• 19:25 - 20:00 - Networking (snacks and drinks)

The event is also supported by the Faculty of Computer Science and Engineering and the Faculty of Electrical Engineering & Information Technologies.

--------------------
More information about the talks and the speakers.

1. Anomaly Detection needs Attention - Sasho Nedelkoski, TU Berlin

Anomaly detection is of great interest to diverse fields and plays a critical role in a wide range of applications, such as in IT Operations, medical health, credit card fraud, and intrusion detection. The beginning of the talk focus on applying the anomaly detection paradigm on such critical applications and discusses an interesting different point of view. The rest of the talk focuses the concept of Artificial Intelligence for IT Operations (AIOps) on planet scale infrastructures. An important part of AIOps platforms is to detect and recognize the anomaly, before it leads to a service or system failure. The talk will discuss anomaly detection on different types of monitoring (log, trace, and metric) data, presenting novel research for parsing and fault detection in textual and sequential data (logs and traces). Then, provides a holistic view of the problem and describes an novel and more general architecture of self-supervised deep generative model for multi-modal anomaly detection, which can largely reduce the amount of false alarms.

Bio:
Sasho Nedelkoski is a doctoral researcher at Berlin Institute of Technology in Germany. He finished his masters at TU Berlin in Computer Science with specialization in Machine Learning and Systems in 2018, and an undergrad in Computer Engineering at FEEIT Skopje.

2. The good, the bad and the ugly in AutoML - Stefan Kochev, Loka.

The DevOps tools for agile software systems development are already to a point of maturity. Continuous development and continuous deployment tools are present in the world of software development for many years. However, these practices need huge workarounds when mixed with Machine Learning pipelines. The problem is a result of the crucial differences between agile software development and the state of how Machine Learning is being practiced. A data science team needs a way to turn their work to production as soon as possible, even without continuous collaboration with DevOps engineers and teams.
A showcase scenario with an automated Machine Learning pipeline, using Kubeflow, would be presented. First, the actual problem we are trying to solve and all the problems which our Machine Learning team faced would be explained. Next we will go through the showcase AutoML pipeline built on top of Kubeflow, Argo Pipelines and Seldon Core. The pros and cons of the proposed AutoML pipeline would be discussed.

Bio:
Stefan Kochev is a software developer at Loka. He is a master’s student at the Faculty of Computer Science and Engineering in Skopje, studying the field of Intelligent Systems. As a newbie in the field of production Machine Learning, he enjoys exploring new tools and platforms aimed to automate the dynamic environment in which the data scientists are solving problems.