Researchers have applied reinforcement learning to improve the way liquidity is provided in decentralized finance (DeFi) markets. They formulated dynamic liquidity provision as a stochastic impulse control problem and used RL to solve it. The learned policies exhibit state-dependent behavior, allocating liquidity based on mispricing, rebalancing costs, uncertainty, inventory exposure, and risk preferences. This approach helps compress the left tail of the profit and loss distribution and avoids catastrophic outcomes under high uncertainty. The results were benchmarked against baseline and sophisticated agents from the AMM microstructure literature. --- Why it matters: This work matters to AI researchers because it demonstrates how reinforcement learning can be applied to real-world problems in finance, such as decentralized markets. The approach could potentially improve market efficiency and reduce risk. Source: https://arxiv.org/abs/2608.19389