Join in zoom Password: 955749
ML for Accelerator Controls: Machine Learning Platform and Generic Optimization Framework and Frontend
The technique of Machine Learning (ML) has been showing promising results in the context of accelerator controls. Together with classical numeric optimization, it has the potential to automate many optimization tasks in the accelerator complex. These tasks tend to be very diverse in the devices they control, their limits, and expert-level details. At the same time, many optimization and ML algorithms have slight differences in their API and making each task compatible with each of them suffers from combinatorial explosion. The BE department is developing two solutions to address these issues: the Machine Learning Platform (MLP) and the Generic Optimization Framework and Frontend (GeOFF).
The MLP is a central platform for storing, versioning and deploying ML models in the CERN Control Center. It allows users to create, update and deploy models with minimal effort, without constraining their workflow or restricting their choice of tools. It also provides tooling to automate seamless model updates as the machine characteristics evolve. Moreover, the system allows model developers to focus on domain-specific development by abstracting infrastructural concerns.
GeOFF is a platform that consists of standardized Common Optimization Interfaces (COI) that harmonize the various optimization algorithms, a generic GUI that applies these algorithms to user-developed optimization tasks, and tools to facilitate the authoring of these tasks.
We present both projects, give examples of previous successful applications, and provide a preview of our next steps in making them more accessible for other working groups at CERN.