Special issue on adaptive and learning agents 2018

Examples of complex application areas that have been successfully tackled by ALA researchers in recent years include energy systems and smart grid, game-playing agents, wind farm control, unmanned aerial vehicles, inventory management and transportation systems. Experimental results in a maze naviga...

Ausführliche Beschreibung

Gespeichert in:
Bibliographische Detailangaben
Veröffentlicht in:Knowledge engineering review 2021, Vol.36, Article e7
Hauptverfasser: Mannion, Patrick, Harutyunyan, Anna, Peng, Bei, Subramanian, Kaushik
Format: Artikel
Sprache:eng
Schlagworte:
Online-Zugang:Volltext
Tags: Tag hinzufügen
Keine Tags, Fügen Sie den ersten Tag hinzu!
Beschreibung
Zusammenfassung:Examples of complex application areas that have been successfully tackled by ALA researchers in recent years include energy systems and smart grid, game-playing agents, wind farm control, unmanned aerial vehicles, inventory management and transportation systems. Experimental results in a maze navigation domain, a coloured flags visiting domain and the Atari game Pong demonstrate that the proposed two-level Q-learning algorithm outperforms the basic Q-learning algorithm and the previously proposed confidence-based HAT algorithm in settings with multiple expert demonstrations and conflicting advice. The authors present a theoretical analysis which proves that TQ-learning converges to a system-efficient equilibrium in the limit, along with empirical results demonstrating that TQ-learning converges to the optimal outcomes in simulated route choice problems. [...]in the paper Fully Distributed Actor-Critic Architecture for Multitask Deep Reinforcement Learning, Valcarcel Macua et al.
ISSN:0269-8889
1469-8005
DOI:10.1017/S0269888921000047