Maya Guillaumont, Project Manager in Capgemini Engineering's R&I Department, explains the progress of her Sustainable AI project. It aims to develop a method for measuring the environmental impact of AI and providing solutions to raise awareness among teams and promote a more eco-responsible use of Artificial Intelligence.
Capgemini Engineering is the Capgemini Group brand that brings together the Engineering and R&D services of Altran, the global industry leader, whose acquisition Capgemini finalized in 2020, with their expertise in digital manufacturing. Capgemini Engineering's in-depth industry knowledge and mastery of cutting-edge digital and software technologies will support the convergence of the physical and digital worlds.
Capgemini Engineering's Research and Innovation Department (with teams in Toulouse, France) researches and develops innovative solutions within the company to offer new products to its clients. Within their projects' framework in Artificial Intelligence, their teams rely on Picsellia to ensure seamless collaboration and better productivity.
The interest in AI has significantly grown in recent years within Capgemini Group. This resulted in the multiplication of the number of AI projects at Capgemini Engineering in all areas: Computer Vision, Natural Language Processing, Machine Learning, etc. But the strong growth in these R&I projects has led to urgent needs for acheiving an optimal organization of the work. Although they have increased, tools are often not adapted to the problems of AI projects (Google Drive, Excel, etc.), collaborative work, and lastly, they take too long to learn.
"We used different tools to manage the project's resources, such as Drive, Neptune and Colab, to centralize our results on Excel for analysis. Each new collaborator had to create an account for each tool and platform, and then we had to give them access to all the resources. With Picsellia, everything can be done in the same place in a centralized way and with a single account.
Collaboration was a huge issue for this project. Because of the constant team evolution, we needed a tool with a quick learning curve to welcome new collaborators more easily and allow workflow automation to ensure reproducibility of results while minimizing human errors related to data handling."
For which missions did you use Picsellia?
The Sustainable AI project team implemented Picsellia for a dozen collaborators to serve as an end-to-end solution, from data management to experiment generation and result analysis.
The project team was able to create a complete working environment, ensuring reproducibility of experiments and continuous monitoring of results, despite the evolution of the project teams.
By integrating Picsellia with the different code execution environments and with some pre or post-processing scripts, the collaborators no longer need to manage several tools.
Thanks to the Datalake and the Experiment Dashboard, for example, all the data is stored in one place, is easy to access, and can be used and versioned automatically at each iteration.
The power of Picsellia lies in the fluidity and agility it allows iterating on the many experiments of an Artificial Intelligence project.
How did your team get used to the platform?
Once the integration is done, it becomes easier to welcome new people on the project who can quickly master the activities and platform.
"I usually do a 5 minutes demonstration of the platform and then the collaborator can start working without worrying about retrieving data sets or saving results since everything is managed by Picsellia."
What about automation for your projects?
The new challenges we are trying to solve are around the automation of the processes related to the experiment planning and execution, to go much further in the completeness of the collected results, but also to allocate the resources to other parts of the project.
To answer this need, we will use a new feature of Picsellia called "Scans" which will allow us to parallelize and automate the experimentation process!
"If you are an enterprise looking to organize your AI work and facilitate collaboration in a simple and modular way, an end-to-end solution from data storage to model deployment like Picsellia is just right for you!"