Citation
Abstract
We present a methodology for partitioning large transformer networks into disconnected components with limited intercommunication, allowing them to be hosted by a collection of machines in a space-based environment. We describe how this method would reduce the memory and power demands of large AI models, outlining a pathway toward their future use in space exploration.
Keywords
neural networks
artificial intelligence
communication network
Details
- Volume
- 42-245
- Published
- May 15, 2026
- Pages
- 1–28
- File Size
- 1.2 MB