Continuous Adaptation via Meta-Learning in Nonstationary and Competitive Environments
This work addresses the challenge of enabling AI agents to learn and adapt quickly in dynamic and adversarial settings, which is incremental but important for advancing towards general intelligence.
The paper tackles the problem of continuous adaptation in nonstationary and competitive environments by developing a gradient-based meta-learning algorithm and testing it in a new multi-agent RoboSumo environment, demonstrating that meta-learning enables significantly more efficient adaptation than reactive baselines in the few-shot regime, with meta-learners outperforming others in competitive scenarios.
Ability to continuously learn and adapt from limited experience in nonstationary environments is an important milestone on the path towards general intelligence. In this paper, we cast the problem of continuous adaptation into the learning-to-learn framework. We develop a simple gradient-based meta-learning algorithm suitable for adaptation in dynamically changing and adversarial scenarios. Additionally, we design a new multi-agent competitive environment, RoboSumo, and define iterated adaptation games for testing various aspects of continuous adaptation strategies. We demonstrate that meta-learning enables significantly more efficient adaptation than reactive baselines in the few-shot regime. Our experiments with a population of agents that learn and compete suggest that meta-learners are the fittest.