For the development of innovative technologies and products, the customer conducts its own research and development (R&D). Experts from the fields of measurement technology, machine learning (ML), frontend and backend software development and user experience form an interdisciplinary team.
Among other things, the customer developed a product with the aim of an automated, location- and device-independent online test procedure for the individual adaptation of medical products. The central aspect of the project is the digitization of the measurement process, which is traditionally carried out on site with the help of specialized equipment and trained personnel. PROTOS Technologie supported the customer in achieving this goal by implementing a modern cloud infrastructure to improve and accelerate the development process.
Machine Learning Operations (MLOps)
When changing from the experimental development phase to production maturity, it became clear to the customer that, above all the Working with Machine Learning Operations (MLOps) brings great benefits. The central prerequisite for the successful implementation of the project was a robust, cost-effective and secure infrastructure which can be continuously tailored and optimized for new machine learning tasks. This means that machine learning engineers can be relieved of the often most time-consuming tasks such as data preparation and model provision and use their capacities for development progress.
In order to integrate new AWS cloud-based MLOps functions into the backend, there had to be close coordination with the respective development teams. The frontend developers were provided with endpoints for uploading data and using the machine learning models with the right interfaces. In cooperation with machine learning engineers and data scientists, a data environment hosted in AWS and a training environment for ML was set up.
Also the Development and maintenance of a scalable CI/CD pipeline, to ensure versioning, verifiability and testing was part of the requirements and interlocks classic DevOps with ML-specific requirements to MLOps.
AWS Machine Learning Pipeline:
PROTOS Technologie supported the customer in the implementation of MLOps so that the necessary infrastructure resources and processes could be made available. Automation has improved, accelerated and secured the development process.
Probably the biggest challenge in machine learning projects is often the enormous amount of time that is required to process data acquisition, data preparation and model training. The biggest advantage of MLOps is the facilitation by automating processes and relieving the ML experts to optimize a machine learning project.
In addition to collection and storage, the large amounts of data required also require prior preparation. The provision on modern cloud infrastructure, as well as the optimization and automation of these factors, with the help of PROTOS’ cloud expertise, made an essential contribution to improving the entire machine learning process.
Implementation in the AWS Cloud:
In the AWS Cloud, a Data Pipeline with AWS Lambda, Amazon S3 and implemented AWS KMS, which encrypts the data, prepares it and supports ongoing data cleaning (data cleaning) and categorization (data labeling).
With the data and the computing capacity provided by AWS, the machine learning model training could be transferred to the AWS cloud in a containerized form. The advantage was that the model training can be run on the scalable and optimized CPUs and GPUs of the cloud, regardless of location and user. In addition, the hosting of the trained models in production could be realized in a lightweight, cost-effective and highly available and scalable manner via AWS Lambda serverless. The possibility of a containerized runtime in AWS Lambda offers modern API-first approaches even for the deployment of complex machine learning models.
To enable interaction with the data pipeline and the ML model results, a secure AWS API Gateway endpoint was made available to mobile and frontend developers.
The highest security standards and intensive testing (unit tests / integration tests) were the basis of the implemented CI/CD process, which automates the deployment and testing of the infrastructure operated on AWS.
All resources were obtained using AWS Cloud Development Kit (CDK) if Infrastructure-as-Code provided. This meant that the complete infrastructure for operating an MLOps-optimized machine learning project could be provisioned and understandably adjusted with one click, and could continue to be operated by the customer himself.