Evolved Policy Gradients

Researchers have released Evolved Policy Gradients (EPG), an experimental metalearning approach that evolves the loss function of AI learning agents to enable faster training on new tasks. The method allows agents to generalize beyond their training data, successfully performing basic tasks in novel scenarios they weren't specifically trained for.

• →EPG is a new metalearning approach that evolves loss functions rather than traditional policy gradient methods.
• →The technique enables AI agents to train faster when encountering novel tasks outside their original training regime.
• →Agents using EPG demonstrated ability to generalize, such as navigating to objects in different locations than seen during training.
• →This represents progress in AI's ability to adapt and generalize beyond specific training scenarios.
• →The research is still experimental, indicating early-stage development of the technology.