The yearly gathering of the merry folks of the OpenMOLE community is coming fast!
We'll be back in late May at the unique site of La Pépinière for a week-long invite-only workshop bringing together the finest pieces of our community.
Twelve researchers and engineers actively collaborating to the project will meet to develop, brainstorm, break, fix, enhance, open new avenues, explore new directions that could benefit the exploration of models at scale.
Here are some (but definitely not all!) of the topics they'll cover during this special week.
Final preparations for the eX Modelo summer school
A new event appears in the OpenMOLE community's diary: the eX Modelo summer school. Aiming to train the new generation of modellers, eX Modelo will welcome a first class of 20 students at the end of June.
For a week, they'll discover the core principles of model exploration, as developed over more than 10 years now by the OpenMOLE community. Attendees can bring their own model along, but will also be able to follow a brand new syllabus especially designed for this first edition.
We're extremely excited with this opportunity to spread the model exploration lore and can't wait to introduce new faces to our vibrant community.
Bridging the gap with the machine learning community
Some tracks of the coding camp have historically set the foundations of future major components of the platform. This was the case a few years back with the inceptions on software container management in the context of scientific pipelines, first achieved during the first coding camp (along with unprecedented prowess involving jQuery and CSS but that's another story...).
This year, we hope to kickstart a new area of focus for the OpenMOLE platform that would make machine learning models first-class citizens in the platform.
We've engaged in a body of research in the field of AutoML (fully/semi-guided design of deep learning model, hyperparameter tuning, ...) that resulted in an experimental framework which delivered promising initial results. The methods studied in AutoTune will, at some point, end up in OpenMOLE, available in combination with our existing methods or as standalone optimisers. The work in this track will focus on iterating over the design and initial implementations of these new methods with distributed computing in mind.
Towards a multi-user deployment of OpenMOLE
Research softwares are not always renowned for their ease of use. We believe that should not be the standard. The OpenMOLE community has, since its early days, been a venue for interdisciplinary collaborations and as such, we highly value the contributions and engagement of users who do not consider themselves computer scientists. It is thus paramount for us to provide them with the easiest user experience, from installation to execution.
But what if they didn't have to install anything on their machine? What if we could just liaise with an IT-savvy super user who could setup a shared instance for the whole lab? We've had this idea in mind for a while and after mingling with a few Kubernetes-based prototypes, we feel confident we are now significantly closer to a multi-user instance of OpenMOLE.
The timing is great, since this new component will be at the heart of our new super secret initiative... Stay tuned and join us in our next jump ;)
Externalise containers management
OpenMOLE has been experimenting with containers since 2016 when we dived deep into the Open Container Image (OCI) specification to generate images compatible with our historical execution engine: PRoot.
Over the years, we've released the details of a generic approach to handle scientific pipelines with PRoot, interfaced with an implementation of the docker engine running in pure user-land, and built custom OpenMOLE tasks to provide a seamless integration of R and Scilab models.
With new execution platforms in mind, we knew we'd have to support different docker backends in the near future. This was the perfect opportunity to start refactoring all the work that had been incorporated to the platform to manage containers, and to modularise it to its own library.
The work on this new Scala package is very advanced and the plan on this front is to put the final touch and remove all the container management code from the main codebase in favour of the new library.
Stay tuned during the week of 20-24 May where we'll send live updates from the front from our @openmole twitter account.