Researchers at OpenAI have developed a new method to reduce variance in policy gradient algorithms. This improvement is achieved by using action-dependent factorized baselines, which are more accurate than traditional baselines. The authors claim that their method can lead to faster and more stable training of reinforcement learning agents. However, the article does not provide experimental results or comparisons with other methods. --- Why it matters: This matters because policy gradient algorithms are widely used in deep reinforcement learning, and reducing variance is crucial for achieving stable and efficient training. This improvement could enable researchers to train more complex models and achieve better performance in tasks such as robotics and game playing. Source: https://openai.com/index/variance-reduction-for-policy-gradient-with-action-dependent-factorized-baselines