Quantum Reinforcement Learning for Aerospace
summary
Hybrid quantum reinforcement learning replaces policy and value neural nets with shallow parameterized circuits and uses a basis embedding, measurement plus linear layers to train a proximal policy optimization agent on the 16-state slippery FrozenLake (80% intended action, 20% slip). Experiments test 19 circuit variants (different entangling topologies and gates) on NISQ-relevant hardware/ simulators and report comparable or better rewards with faster convergence and fewer environment interactions versus similarly sized classical nets. Tried metrics — normalized effective dimension, entanglement capability, expressibility — and find no clear correlation with performance.
AI-written summary · about
chapters (14)
- 0:00Intro
- 0:40Presentation
- 1:56Airbus
- 4:10Why Quantum
- 5:12Airbus Quantum Challenge
- 6:35Quantum Use Cases
- 7:22Quantum Optimization Cluster
- 8:13Quantum Flight Path Optimization
- 10:49Project Overview
- 11:25Another Use Case
- 12:55Motivation
- 14:05Environment
- 16:38Results
- 18:10Conclusion
related talks
31mReinforcement Learning with Quantum Algorithms: Simulating Nematode Behaviour
19mQuarl: A Learning-Based Quantum Circuit Optimizer
19mReinforcement Learning for Optimizing Hybrid Quantum-Classical Workflows
15mEnsuring Robustness in Hybrid Quantum Neural Networks
32mQuantum Artificial Intelligence
15mAccelerQ: Accelerating Quantum Eigensolvers With Machine Learning on Quantum Simulators
26mBeyond the Hype: Hybrid Quantum-Classical AI Agents for Scientific Discovery
31mQuantum neural networks