Reinforcement Learning for Optimizing Hybrid Quantum-Classical Workflows
summary
Hybrid workflow links DFT outputs from CP2K → Qiskit integration to build VQE/UCCSD circuits for corrosion inhibitors on periodic aluminum; circuits span 150 to 13,700 gates (depths 83–9,700). A brute‑force pruning drops a UCCSD circuit from 150 to 126 gates (16% reduction), and a reinforcement‑learning agent that removes or replaces gates achieves good results on H2 (reduced circuit ~17 gates, mean final reward 1.9) but stalls on LiH due to an ~11,500‑action space; simulations run on statevector and H2 on IQM, with plans for IQM Garnet and IonQ Aria1 plus error mitigation.
AI-written summary · about
chapters (13)
- 0:00Introduction
- 1:04Company and quantum efforts
- 2:09Corrosion inhibition materials
- 3:15Hybrid workflow for simulation
- 4:46Benchmarking VQE circuits
- 5:50Hardware constraints
- 6:53Circuit size reduction
- 7:54Brute force gate reduction
- 9:19RL for circuit optimization
- 11:12Refining RL reward scheme
- 12:16RL optimization results
- 13:45Conclusion and future work
- 15:53Q&A
related talks
21mQuantum Reinforcement Learning for Aerospace
19mExtracting maximum value from today's quantum computers while prep
42mBuilding Hybrid Quantum-Classical Pipelines : A Practic...
19mQuarl: A Learning-Based Quantum Circuit Optimizer
32mComputational quantum chemistry: Classical gaps, quantum opportunities, open questions Part 2
41mQuantum Computing in Practice: Where We Are, What Works, and What’s Next
15mAccelerQ: Accelerating Quantum Eigensolvers With Machine Learning on Quantum Simulators
27mA Haskell Adiabatic DSL: Solving Classical Optimization Problems on Quantum Hardware