ML-TOSCA: ML pipeline modelling and orchestration using TOSCA
Date
2023
Authors
Journal Title
Journal ISSN
Volume Title
Publisher
Tartu Ülikool
Abstract
In today’s world, machine learning is increasingly involved in different areas. Moreover,
automating machine learning workflows through AutoML enables organizations to
develop and deploy machine learning solutions at scale rapidly. Additionally, leveraging
the power of cloud computing can provide even greater scalability and flexibility, allowing
us to efficiently process large datasets and cost-effectively train and implement complex
machine learning models. Undoubtedly, these technologies will play an essential role in
shaping the future across various industries. Despite many advantages, there is a lack
of widespread combined implementations of AutoML and cloud-based solutions. This
thesis describes a new AutoML integration approach to the TOSCA standard. TOSCA
is an open-source specification used to describe the topology of cloud applications and
services. Incorporating AutoML techniques into TOSCA enables users to automatically
generate optimized machine learning models with the help of cloud applications, which
can improve the speed and efficiency of model creation. The proposed approach is
implemented in the RADON ecosystem, allowing node and relationship types to be
created. The final solution allows users to create and join blocks to define a complete
machine learning pipeline structure.
Description
Keywords
AutoML, TOSCA, ML-TOSCA, ML pipeline, Pipeline