Would it be possible/practical to build a distributed deep learning engine by tapping into ordinary PCs' unused resources?

I started thinking about this in the context of Apple's new line of desktop CPUs with dedicated neural engines. From what I hear, these chips are quite adept at solving deep learning problems (as the name would imply). Since I can only imagine the average user wouldn't necessarily be optimizing cost functions on a regular basis, I was wondering if it would be theoretically possible to use those extra resources set up some type of distributed network similar to a spark cluster, only instead of having 5 to 20 dedicated systems, you would have maybe 200 to 300 nodes consisting of whatever portion of available resources the user decided to rent out. I'm imagining an economic arrangement somewhat similar to a crypto mining pool wherein the owner of each node gets paid according to the extent of their contribution.

Topic apache-spark deep-learning map-reduce machine-learning

Category Data Science

1
answered at

I doubt that this could work:

  • it's not reliable for the customer, since the owners of the machines can switch it off or start playing some video-intensive game whenever they want.
  • in order to make it interesting enough for the owners the price would have to be substantial, otherwise it's not even worth the constraint. So it's very unlikely to be competitive compared to commercial cloud providers, and this is a fairly competitive market: these providers buy large servers at the best price for the computing power, and they have a very efficient infrastructure.
  • There's an obvious issue with privacy, so everything would have to be encrypted and this adds an overhead in processing time. Also some owners might be worried about potential flaws which would let the customer hack their computer.

About

Geeks Mental is a community that publishes articles and tutorials about Web, Android, Data Science, new techniques and Linux security.