Quantum reinforcement learning - Quantum PhysicsReportar como inadecuado




Quantum reinforcement learning - Quantum Physics - Descarga este documento en PDF. Documentación en PDF para descargar gratis. Disponible también para leer online.

Abstract: The key approaches for machine learning, especially learning in unknownprobabilistic environments are new representations and computation mechanisms.In this paper, a novel quantum reinforcement learning QRL method is proposedby combining quantum theory and reinforcement learning RL. Inspired by thestate superposition principle and quantum parallelism, a framework of valueupdating algorithm is introduced. The state action in traditional RL isidentified as the eigen state eigen action in QRL. The state action set canbe represented with a quantum superposition state and the eigen state eigenaction can be obtained by randomly observing the simulated quantum stateaccording to the collapse postulate of quantum measurement. The probability ofthe eigen action is determined by the probability amplitude, which isparallelly updated according to rewards. Some related characteristics of QRLsuch as convergence, optimality and balancing between exploration andexploitation are also analyzed, which shows that this approach makes a goodtradeoff between exploration and exploitation using the probability amplitudeand can speed up learning through the quantum parallelism. To evaluate theperformance and practicability of QRL, several simulated experiments are givenand the results demonstrate the effectiveness and superiority of QRL algorithmfor some complex problems. The present work is also an effective exploration onthe application of quantum computation to artificial intelligence.



Autor: Daoyi Dong, Chunlin Chen, Hanxiong Li, Tzyh-Jong Tarn

Fuente: https://arxiv.org/







Documentos relacionados