Auto.gov: Learning-based Governance for Decentralized Finance (DeFi)

Traditional DeFi governance relies on manual parameter adjustments prone to human bias and financial risks. This research introduces Auto.gov, a learning-based governance framework using deep Q-network reinforcement learning for semi-automated, data-driven parameter adjustments. 

‍

Testing on real-world data shows Auto.gov outperforms benchmark approaches by at least 14% and static baseline models by tenfold in protocol profitability, whilst demonstrating capability to retain funds that would otherwise be lost to price oracle attacks.

‍