Can Blockchain Make Distributed AI Training Work?

Proof of Work asks thousands of computers to solve a puzzle that has no value outside the protocol. The winner gets a reward, everyone else’s work is thrown away, and the cycle repeats. The computation is wasted, but the system is very robust. No central authority, no trusted party, no coordination beyond following the rules.

But what if we replaced the hash puzzle with something useful, like training a neural network?

This idea has a big problem: verifying that a gradient update is honest is much harder than verifying a hash. But there is a more interesting question behind it. Can the structure of a blockchain (independent nodes, local decisions, emergent agreement) serve as a foundation for distributed machine learning? We think it can, and we made a small experiment that supports it.

The core idea

Each node holds a copy of a neural network. Nodes compete to find good...

Copyright of this story solely belongs to hackernoon.com. To see the full text click HERE