Dr. Larysa Visengeriyeva

The hype around Machine Learning and Artificial Intelligence can give the illusion that software with integrated ML models is easy to develop. However, according to various studies, almost 80% of ML projects fail to get into production. In this talk, I will present 10 fundamental practices for machine learning engineering to succeed in your own project.

An important reason why companies fail to implement artificial intelligence and/or machine learning is the difficulty of identifying meaningful use cases for machine learning with a shared understanding between domain experts, ML specialists, data scientists, and developers.

In this hands-on workshop, we will demonstrate how we can use ideas from Domain-Driven Design, Collaborative Modelling and Canvasses to develop a common understanding of our product, identify AI/ML Use Cases for innovation, and structure a machine learning project

If you want to develop good, innovative, and data-driven software products, you should not start by evaluating machine learning algorithms. The first step should be to find and verify an AI/ML Use Case so that the use of AI/ML will solve a real problem. However, the whole process from the identification of the use case to the introduction of ML models in the company is not a trivial procedure.

In this hands-on workshop, we will talk a little bit about machine learning basics and then we will leverage techniques like EventStorming and the ML Design Canvas. Event Storming is a method of Collaborative Modeling that helps technical experts, developers and all other project participants* to develop a common understanding of a business domain and thus identify possible use cases for innovative AI/ML technologies. Each potential use case is then formulated as an ML problem using the ML Design Canvas. Furthermore, the ML Design Canvas is used to structure the ML project and specify all components. We will also draw parallels to the Bounded Context Design Canvas and show how this approach fits into approaches like the model exploration whirlpool and/or the DDD Crew (GitHub.com/ddd-crew) starter modeling process.

In diesem Vortrag stellt Larysa Visengeriyeva fundamentale Praktiken für Machine Learning Engineering vor, die dabei helfen, das eigene ML Projekt zum Erfolg zu bringen.

Ein wichtiger Grund, warum Unternehmen an der Umsetzung von AI/ML scheitern, ist die Schwierigkeit, einen sinnvollen Use Case für ML zu identifizieren. Bei INNOQ haben wir einen Konzept für die Erkennung von ML Use Cases entwickelt und stellen unseren Prozess in diesem Vortrag vor. In den ersten zwei Phasen nutzen wir DDD Methoden wie EventStorming, um die Fachlichkeiten zu verstehen und potenzielle ML Tasks zu identifizieren. Das Ziel der zweiten Phase ist das ML-Projekt zu strukturieren und die Anforderungen zu klären. Dafür nutzen wir ein visuelles Framework: das ML Design Canvas. Dabei werden wichtige funktionale und qualitative Anforderungen für das ML-System spezifiziert.