Two years ago, we laid out our vision for a machine learning compute protocol. One that connects every device in the world into an open network for machine intelligence, with no gatekeepers or artificial boundaries.
This week, we’re sharing more of our progress in public, beginning with solutions to some of the core challenges:
• Consistent Execution, ensuring compatibility across any device; • Trustless Verification, checking and agreeing on work performed in a scalable way; and • Efficient Communication, sharing workloads between devices over the internet.
Next month, we’ll release our Testnet, bringing these components together into a single network for open machine learning - accessible to humans and machines alike and scalable to every computer in the world.
Select from the tags below to view related links.
Execution
RL Swarm
A peer-to-peer system for collaborative reinforcement learning over the internet, accessible by anyone on consumer or datacentre hardware.
An arbitration system for machine learning over untrusted nodes, enabled by RepOps, a library for bitwise reproducibility across heterogeneous devices.
A framework for training mixture-of-experts models in a parallel fashion, showing that introducing heterogeneity into the training process yields a better-performing ensemble.
