AccelerQ: Accelerating Quantum Eigensolvers With Machine Learning on Quantum Simulators
summary
A framework automatically tunes variational quantum eigensolver implementations by treating them as black-box programs, training an XGBoost regressor on classically simulable systems up to 16-qubit classical training, then applying learned hyperparameter suggestions plus genetic-algorithm search on 20–28 qubit Hamiltonians. Results show concrete error drops: ADAPT-QSCI: 5.48%→5.05% (5.3% with ML alone, 5.05% with GA) and QCELS: 7.5%→6.5% (no GA improvement); some 20–24 qubit cases are inconclusive and full evaluations take months.
AI-written summary · about
chapters (10)
- 0:00AccelerQ and eigensolvers
- 1:27Optimizing quantum software
- 2:56Parameter tuning challenges
- 4:03Quantum code abstraction
- 6:32ML model training
- 8:20Greedy parameter selection
- 10:09Parameter relationships
- 11:53Validating constraints
- 12:56Results comparison
- 13:58Q&A
related talks
47mQuantum Machine Learning: Cutting Through the Buzzword BS
21mCUDA-QX and Artificial Intelligence for Quantum Computing
01h 36mHands on Quantum Computer
19mExploring Quantum Advantage for the German industry
01h 06mPractical Quantum Machine Learning: Overcoming the Bottlenecks of Scalability and(…)
29mA Rigorous and Robust Quantum Speed-up in Supervised Machine Learning
44mQuantum machine learning: Prospects and challenges
19mQuantum Computing for Simulation in Manufacturing