PoLChain
Proof-of-Learning Protocol — Decentralized AI Training on Base Sepolia
4MINERS
3SMART CONTRACTS
Base SepoliaTESTNET · CHAINID 84532
RESEARCH PROTOTYPE
OPENAI 2024$3.0B training cost vs $3.7B revenue
ANTHROPIC 2024$1.5B training cost vs $2.55B revenue
Training costs consumed 81% of OpenAI's revenue and 59% of Anthropic's in 2024.
THE PROTOCOL ANSWER
Distributed compute paid in protocol tokens.
Cryptographic proof of training, not just claims.
Federated aggregation produces a shared model as a side effect.
Block lifecycle
The centerpiece flowTask PostedReward escrowed in TaskManager.
Miners TrainEach of 4 miners trains MNISTNet locally.
ZK Proof GeneratedEZKL + Halo2 prove the forward pass.
Proof Submitted On-ChainsubmitWork invokes the Halo2Verifier.
Winner SelectedfinalizeTask picks the highest score.
FedAvg AppliedWinning shard merges into global model.
$POL RewardedEscrowed POL transferred to winner.