Advances in Reinforcement Learning

The field of Reinforcement Learning (RL) is rapidly evolving, with a growing focus on addressing non-Markovian dynamics and improving the efficiency of decision-making processes. Researchers are exploring new methodologies to handle complex temporal correlations and develop more flexible and adaptive algorithms. Notably, the integration of contextual bandits and deep RL is showing promise in enhancing policy flexibility and computational efficiency. Furthermore, the incorporation of fractional calculus and novel action duration selection methods are providing new avenues for improving the performance of RL systems. One of the key areas of innovation is the development of more sophisticated RL algorithms, such as those utilizing Double Q-learning and Fractional Policy Gradients, which are demonstrating significant improvements over traditional methods. Some noteworthy papers in this area include:

  • Constructing Non-Markovian Decision Process via History Aggregator, which introduces a novel methodology for addressing non-Markovian dynamics.
  • Fractional Policy Gradients: Reinforcement Learning with Long-Term Memory, which proposes a framework incorporating fractional calculus for long-term temporal modeling.
  • Double Q-learning for Value-based Deep Reinforcement Learning, Revisited, which adapts the core idea of Double Q-learning for value-based deep RL and demonstrates reduced overestimation and improved performance.

Sources

Constructing Non-Markovian Decision Process via History Aggregator

Novel RL approach for efficient Elevator Group Control Systems

Adaptive Action Duration with Contextual Bandits for Deep Reinforcement Learning in Dynamic Environments

Fractional Policy Gradients: Reinforcement Learning with Long-Term Memory

Double Q-learning for Value-based Deep Reinforcement Learning, Revisited

Built with on top of