$\mathcal{B}$-Coder: Value-Based Deep Reinforcement Learning for Program Synthesis at ICLR 2024

Program synthesis benefits from value-based RL methods, showcasing stability and performance improvements.