As an aerospace engineer, I have limited expertise in the field of computing. However, I am currently working on a project where we aim to control the power consumption of a data center. During my research, I came across federated learning and it seems like a promising solution for our goal.
We plan to build small-scale data centers for training AI models and we believe that by controlling the number of calculations performed by the servers, we can directly control the power consumption. To achieve this, we need computational tasks that are small, scalable, and interruptible without wasting resources. Parallel computing seems to be a good fit for these requirements as it allows us to interrupt the computations without affecting the overall progress of the computation.
Federated learning also aligns well with our goals as it is a distributed computing architecture where computers work independently on larger tasks and can continue to operate without disruption even if one of the computers stops performing calculations.
Our objective is to build a proof-of-concept that demonstrates our ability to accurately control the energy consumption of our computers. I would greatly appreciate any advice on where to start. Can I install software on some old servers and participate in federated learning? Are there any ideas on how I can contribute to federated learning while still having control over the number of calculations performed?
Thank you in advance for your help!