Symbolic Task Inference in Deep Reinforcement Learning (Abstract Reprint)
Proceedings of the AAAI Conference on Artificial Intelligence · 14.03.2026 · 40 (47) · 39867-39867
DOI: 10.1609/aaai.v40i47.41382 ↗Kurz gefasst
Fachartikel in der Zeitschrift Proceedings of the AAAI Conference on Artificial Intelligence, online erschienen am 14.03.2026. Verfasst von Hosein Hasanbeig und 4 weiteren Autor:innen.
Abstract (Originalsprache)
This paper proposes DeepSynth, a method for effective training of deep reinforcement learning agents when the reward is sparse or non-Markovian, but at the same time progress towards the reward requires achieving an unknown sequence of high-level objectives. Our method employs a novel algorithm for synthesis of compact finite state automata to uncover this sequential structure automatically. We synthesise a human-interpretable automaton from trace data collected by exploring the environment. The state space of the environment is then enriched with the synthesised automaton, so that the generation of a control policy by deep reinforcement learning is guided by the discovered structure encoded in the automaton. The proposed approach is able to cope with both high-dimensional, low-level features and unknown sparse or non-Markovian rewards. We have evaluated DeepSynth’s performance in a set of experiments that includes the Atari game Montezuma’s Revenge, known to be challenging. Compared to approaches that rely solely on deep reinforcement learning, we obtain a reduction of two orders of magnitude in the iterations required for policy synthesis, and a significant improvement in scalability.
Zitieren
APA
Hasanbeig, H., Jeppu, N. Y., Abate, A., Melham, T., & Kroening, D. (2026). Symbolic Task Inference in Deep Reinforcement Learning (Abstract Reprint). Proceedings of the AAAI Conference on Artificial Intelligence, 40(47), 39867-39867. https://doi.org/10.1609/aaai.v40i47.41382
Vancouver
Hasanbeig H, Jeppu NY, Abate A, Melham T, Kroening D. Symbolic Task Inference in Deep Reinforcement Learning (Abstract Reprint). Proceedings of the AAAI Conference on Artificial Intelligence. 2026;40(47):39867-39867. doi:10.1609/aaai.v40i47.41382
BibTeX
@article{Hasanbeig2026,
author = {Hasanbeig, Hosein and Jeppu, Natasha Yogananda and Abate, Alessandro and Melham, Tom and Kroening, Daniel},
title = {Symbolic Task Inference in Deep Reinforcement Learning (Abstract Reprint)},
journal = {Proceedings of the AAAI Conference on Artificial Intelligence},
year = {2026},
volume = {40},
number = {47},
pages = {39867--39867},
doi = {10.1609/aaai.v40i47.41382}
}
EUVIMED