CLAIMED, a visual and scalable component library for Trusted AI

Deep Learning models are getting more and more popular but constraints on explainability, adversarial robustness and fairness are often major concerns for production deployment. Although the open source ecosystem is abundant on addressing those concerns, fully integrated, end-to-end systems are lacking in open source. Therefore we provide an entirely open source, reusable component framework, visual editor and execution engine for production grade machine learning on top of Kubernetes, a joint effort between IBM and the University Hospital Basel. It uses Kubeflow Pipelines, the AI Explainability360 toolkit, the AI Fairness360 toolkit and the Adversarial Robustness Toolkit on top of ElyraAI, Kubeflow, Kubernetes and JupyterLab. Using the Elyra pipeline editor, AI pipelines can be developed visually with a set of jupyter notebooks. We explain how we’ve created a COVID-19 deep learning classification pipeline based on CT scans. We use the toolkit to highlight parts of the images which have been crucial for the models decisions. We detect bias against age and gender and finally, show how to deploy the model to KFServing to share it across different hospital data centers of the Swiss Personalized Health Network. Open source software for performing individual AI pipeline tasks are abundant, but the community lacks a fully integrated, trusted and scalable visual tool. Therefore we have built CLAIMED, the visual Component Library for AI, Machine Learning, ETL and Data Science which runs on top of ElyraAI capable of pushing AI pipelines of any kind to Kubernetes. Any containerized application can become a component of the library. CLAIMED has been released 3 under the Apache v2 open source license. In the following we introduce the open source components we are integrating in our current release, followed by an overview of different component categories paired with a description of exemplary components used in health care. This pipeline is also available in open source.