Seen a paper on Proof-Of-Learning and wanted to know-how the consensus works and how to implement it in ethereum?
I will try and summarize this for you shortly:
So currently POL, Proof Of Learning, is supported on a coin named WekaCoin. And POL is based on machine learning.
The WekaCoin nodes, called trainers, also operate in a peer-to-peer network just as in most blockchains, but instead of submitting a nonce for example, they submit machine learning models for specific layed out tasks that were previously published by other nodes in the network, called suppliers. (This is then their submitted "answers" to the "block problem".) Note that these models are executed on data that was not observed by the trainers during training and ranked in a distributed fashion according to a performance metric determined by the task supplier.
The trainers, which can be compared to validators on Ethereum or miners on Bitcoin, are randomly selected nodes of the network in charge of ranking models and proposing new blocks to the chain. The trainer owning the best model is rewarded by the supplier with a certain amount of WekaCoins.
The trainers are also evenly compensated with a transaction fee paid by the task supplier as well as with new WekaCoins.
"This whole process is used to validate blocks of transactions in a blockchain. The resulting blockchain is a distributed and veriﬁable database of transactions plus an open repository of machine learning models and experiments used to verify them.
Proof-of-Learning establishes a symbiotic relationship be- tween two complex and unrelated tasks: 1) validate transac- tions in a distributed ledger, and 2) storing machine learning models and experiments in a distributed database.
The idea of aligning two different tasks is inspired by reCAPTCHA , a mechanism that allows detecting whether a web user is human (for security reasons) while assisting to digitize printed material."