The promises of Machine Learning (ML)-aided medicine have yet to be realized, in large part because robust optimization and generalizability of ML models requires large training datasets, which are logistically challenging to obtain. The majority of ML medical research is conducted using small- to mid-scale single institution datasets and does not equitably represent the general population. For institutions with smaller footprint, resources, and catchment areas, the problem is exacerbated, making engagement in high-quality ML model development challenging. The emergence of Federated Learning (FL) as a new discipline has great potential to empower smaller institutions to collaboratively train algorithms for healthcare applications without sharing private patient data. This will open the door for application of ML-assisted medicine in rural settings and other underrepresented contexts, including for conditions with low prevalence. However, there remain numerous pragmatic concerns of how multicenter collaborations can be coordinated between small, similarly resourced groups outside of joining large consortiums that are coordinated around centralized, well-resourced leadership. We seek to address this problem by discussing the challenges faced by federated coalitions of smaller biomedical research institutions and proposing a three-stage plan to facilitate (1) the formation of an inter-institutional Working Group, (2) the design and engineering of operating policies and infrastructure, and (3) the execution of study and open dissemination of lessons learned.