Privacy-Focused Distributed Computing for LLMs/DL/ML?

Question

I'm exploring the idea of a distributed computing platform that enables fine-tuning and inference of LLMs and classical ML/DL using computing nodes like MacBooks, desktop GPUs, and clusters.The key differentiator is that data never leaves the nodes, ensuring privacy, compliance, and significantly lower infrastructure costs than cloud providers. This approach could scale across industries like healthcare, finance, and research, where data security is critical.I would love to hear honest feedback. Does this have a viable market? What are the biggest hurdles?

blackbear_ · Accepted Answer

Sounds like federated learning [1]. There is lots of research on that, and therefore a lot of interest too. This is not my field so I will just point to some surveys [2,3] to get you started.
[1]: https://arxiv.org/abs/2211.08413
[2]: https://arxiv.org/pdf/2009.13012
[3]: https://arxiv.org/pdf/2111.08834